Zero-Sum Multinomial Distribution

Zero-Sum Multinomial Distribution—commonly abbreviated ZSM—is the stationary probability distribution of species relative abundances within a dispersal-limited local ecological community under Stephen P. Hubbell's (2001) Unified Neutral Theory of Biodiversity and Biogeography (UNTBB), solved analytically in closed integral form by Igor Volkov, Jayanth R. Banavar, Stephen P. Hubbell, and Amos Maritan (2003, Nature) and Alan J. McKane, David Alonso, and Ricard V. Solé (2000, 2004). In Hubbell's neutral framework, all individuals across all trophically similar species in an ecosystem are assumed to possess identical per-capita birth, death, dispersal, and speciation rates (equivalent to setting niche stabilization 1 − ρ = 0 and fitness ratio κ₂ / κ₁ = 1 in stabilizing and equalizing coexistence mechanisms). The ZSM emerges from a two-scale metacommunity coupling: a local community saturated at a fixed total number of individuals J_L obeys an ecological zero-sum game (where every death of an individual creates a vacancy filled either with probability 1 − m by the local offspring of a resident individual or with probability m by an immigrant propagule drawn from the surrounding regional metacommunity), while the regional metacommunity of size J_M ≫ J_L evolves under neutral drift and point-mutation speciation (rate ν) toward R. A. Fisher's logseries equilibrium governed by the dimensionless fundamental biodiversity number θ = 2 J_M ν (or J_M ν / (1 − ν)). Whereas complete dispersal (m = 1) causes the local community to mirror the metacommunity's Fisher logseries distribution, any finite dispersal limitation (0 < m < 1) depletes the rarest singleton and doubleton classes locally, generating a left-skewed (rare-species-deficient) curve on Preston's log₂ abundance octaves that provides close quantitative fits to hyperdiverse tropical rainforest tree plots (such as the 50-hectare Barro Colorado Island plot in Panama), coral reefs, and planktonic foraminifera assemblages.

Ecological Neutrality and the Per-Individual Equivalence Postulate

Inspired by Motoo Kimura's neutral theory of molecular evolution (1968, 1983) in population genetics and Robert H. MacArthur and Edward O. Wilson's equilibrium theory of island biogeography (1967), Hubbell (1979, 1997, 2001) formulated the Unified Neutral Theory of Biodiversity and Biogeography on two foundational postulates:

1. Per-Capita Demographic Equivalence: Every individual belonging to a guild of trophically similar, competing species (such as all canopy and understory trees in a closed-canopy moist tropical forest, or all hermatypic corals on a reef crest) has the same per-capita probability of giving birth, dying, immigrating, and speciating, regardless of species identity.
2. The Ecological Zero-Sum Game (J_L = Constant): Because limiting space, light, water, or energy saturates total guild biomass (connected conceptually to the more-individuals hypothesis and mass-ratio hypothesis), the total number of individuals summed across all S species in a local community of area A remains constant at J_L = ∑_i=1^S n_i:

ΔJ_L = ∑_i=1^S Δn_i = 0 Consequently, no species can increase in local abundance (n_i → n_i + 1) without another species simultaneously losing an individual (n_j → n_j − 1).

Coupling the Metacommunity (θ) to the Local Community (J_L, m)

To account for dispersal limitation across spatial scales, Hubbell partitioned ecological dynamics into two hierarchical levels:

• The Panmictic Metacommunity (J_M, ν, θ): On evolutionary timescales, the regional source pool of J_M individuals undergoes a neutral Moran birth–death process with point-mutation speciation at per-birth rate ν (where J_M ≫ J_L and ν ≪ 1). By mathematical isomorphism with Watterson's (1974) infinite-alleles neutral model and Ewens's (1972) sampling formula in population genetics, the stationary distribution of species abundances in the metacommunity is fully determined by a single compound parameter, the fundamental biodiversity number:

θ = 2 J_M ν (for continuous-time overlapping generations, or J_M ν / (1 − ν) in discrete replacements). In the continuum limit (J_M → ∞ with θ finite), the expected number of metacommunity species having relative abundance inside the interval (x, x + dx) follows Fisher's logseries density: f_meta(x) dx = θ ( (1 − x)^θ−1 / x ) dx

• The Dispersal-Limited Local Community (J_L, m): On ecological timescales, speciation inside the local plot of J_L individuals is negligible. Instead, whenever one of the J_L local individuals dies, the vacated site is colonized via two competing pathways:
  1. With probability 1 − m, the vacancy is filled by a seed or larva produced within the local community, chosen with probability n_i / (J_L − 1) proportional to each species' current local count;
  2. With probability m (where 0 < m ≤ 1 is the immigration probability), the vacancy is filled by an immigrant propagule arriving from the regional metacommunity, belonging to species i with probability x_i equal to that species' relative abundance in the metacommunity.

Birth–Death Transition Rates for Local Species Abundance (n)

Following McKane, Alonso, and Solé (2000) and Volkov et al. (2003), consider a focal species i whose relative abundance in the surrounding metacommunity is x ∈ (0, 1) and whose current integer abundance in the local community of size J_L is n ∈ {0, 1, …, J_L}. Because all non-focal species collectively have local abundance J_L − n and metacommunity relative abundance 1 − x, the zero-sum dynamics of species i reduce to a one-step master equation (birth–death Markov chain) on the state space n ∈ {0, 1, …, J_L}:

• Effective Local Birth Rate b(n | x) (Transition n → n + 1): For species i to gain one individual, one of the J_L − n individuals belonging to other species must die (probability (J_L − n) / J_L), and the resulting vacancy must be replaced either by a local parent of species i (probability (1 − m) n / (J_L − 1)) or by an immigrant of species i from the metacommunity (probability m x):

b(n | x) = ( (J_L − n) / J_L ) [ (1 − m) ( n / (J_L − 1) ) + m x ]

• Effective Local Death Rate d(n | x) (Transition n → n − 1): For species i to lose one individual, one of its own n individuals must die (probability n / J_L), and the vacancy must be filled either by a local parent of another species (probability (1 − m) (J_L − n) / (J_L − 1)) or by an immigrant of another species (probability m (1 − x)):

d(n | x) = ( n / J_L ) [ (1 − m) ( (J_L − n) / (J_L − 1) ) + m (1 − x) ]

Detailed Balance and the Polya-Urn / Beta-Binomial Conditional Equilibrium

Because a one-dimensional birth–death chain on a finite state space satisfies detailed balance at stationarity, the conditional stationary probability P(n | x) that a species with metacommunity frequency x has n individuals in the local community obeys: P(n | x) b(n | x) = P(n + 1 | x) d(n + 1 | x) Iterating from n = 0 gives the product solution: P(n | x) = P(0 | x) ∏_k=0ⁿ⁻¹ [ b(k | x) / d(k + 1 | x) ]

Defining the composite dispersal-coupling parameter γ (which measures the effective number of immigrants entering the local community per local turnover cycle): γ = m (J_L − 1) / (1 − m) the conditional stationary distribution P(n | x) simplifies to a closed-form Polya-urn (beta-binomial) distribution expressed via Gamma functions Γ(·) or Pochhammer rising factorials (a)_n = Γ(a + n) / Γ(a): P(n | x) = C(J_L, n) [ (γ x)_n (γ (1 − x))_{JL − n} ] / (γ)_JL where C(J_L, n) = J_L! / [ n! (J_L − n)! ] is the binomial coefficient.

Integration Over the Metacommunity to Obtain the ZSM Formula

Finally, Volkov et al. (2003) obtained the unconditional expected number of species ⟨ϕ_n⟩ containing n individuals in the local community (for 1 ≤ n ≤ J_L) by integrating the conditional Polya-urn probability P(n | x) against the metacommunity Fisher logseries distribution f_meta(x) = θ (1 − x)^θ−1 / x: ⟨ϕ_n⟩ = ∫₀¹ P(n | x) f_meta(x) dx = θ ( J_L! / [ n! (J_L − n)! ] ) ( Γ(γ) / Γ(J_L + γ) ) ∫₀¹ [ Γ(n + γ y) / Γ(1 + γ y) ] [ Γ(J_L − n + γ (1 − y)) / Γ(γ (1 − y)) ] (1 − y)^θ−1 dy where the integration variable y = x runs from 0 to 1. Subsequently, David Alonso, Rampal S. Etienne, and Alan J. McKane (2006) and Etienne (2005) derived a genealogical coalescent sampling formula (generalizing Ewens's sampling formula to two spatial scales via ancestral lineage coalescence and immigration parameter I = γ = m(J_L − 1)/(1 − m)) that allows rapid maximum-likelihood estimation of (θ, m) without numerical integration.

Asymptotic Limits at m = 1 (Fisher Logseries) and m → 0 (Monodominance)

The analytical ZSM formula illuminates how dispersal limitation transforms community structure across the range 0 < m ≤ 1:

1. Panmictic / Unlimited-Dispersal Limit (m → 1, so γ → ∞): When every local death is replaced directly by a random propagule drawn from the metacommunity (m = 1), local birth–death drift disappears and the ZSM collapses to Fisher's logseries distribution with parameter α = θ:

⟨ϕ_n⟩ → θ ( x_Lⁿ / n ) Under m = 1, singletons (n = 1) are always the most speciose abundance class, and a Preston octave plot declines monotonically from left to right.

1. Strong Dispersal-Limitation Limit (m → 0, so γ → 0): As immigration m becomes small, rare immigrants entering at n = 1 face a high probability of drifting to local extinction (n = 0) before reproducing, whereas resident species that happen to reach moderate abundance self-reinforce via local reproduction (1 − m) n / (J_L − 1). Consequently, the singleton class (n = 1) and lowest octaves are depleted relative to intermediate octaves, creating an interior modal peak and a pronounced left-skew (excess of rare species relative to a symmetric Gaussian lognormal, but deficit of singletons relative to Fisher's logseries). In the limit m → 0, local richness collapses to a single monodominant species (n = J_L).

Empirical Performance on Tropical Forests and Coral Reefs

In Hubbell (2001) and Volkov et al. (2003, 2007), the two-parameter ZSM distribution (θ, m) was tested against census data from the Center for Tropical Forest Science (CTFS) global network of 50-hectare forest dynamics plots—including Barro Colorado Island (BCI), Panama (J_L = 21,457 trees with trunk diameter ≥ 10 cm across S = 225 species, fitted with θ ≈ 47.7 and m ≈ 0.10), Pasoh Forest Reserve in Malaysia (S = 678 species, θ ≈ 186, m ≈ 0.09), Yasuní in Ecuador, and Korup in Cameroon—as well as Indo-Pacific coral-reef transects:

• Across all tested tropical plots, the ZSM captured the observed asymmetric left-skew of Preston log₂ octaves, outperforming both the one-parameter Fisher logseries (which overpredicts singletons in dispersal-limited plots) and the two-parameter canonical lognormal (which cannot fit the long, asymmetric left tail of rare species without truncating at a veil line).

Scientific Critiques, Non-Identifiability, and Niche–Neutral Syntheses

While the empirical performance of the ZSM stimulated a renaissance in quantitative macroecology, subsequent theoretical work highlighted important mechanistic caveats:

1. Process-to-Pattern Non-Identifiability: Purves and Pacala (2005), Chave (2004), and McGill et al. (2006) demonstrated that static species abundance distributions (including the ZSM curve) can also be generated by non-neutral models such as stochastic niche theory (Tilman, 2004), hierarchical niche partitioning, or pulliam source–sink dynamics, meaning that a close statistical fit to the ZSM does not by itself prove that niche differences (1 − ρ) are zero.
2. Species Lifespans and Turnover Rates: Ricklefs (2006) and Nee (2005) noted that pure neutral drift requires very long times (millions of generations) for a newly speciated taxon to drift from n = 1 to widespread canopy dominance, motivating modern continuum models (Adler, HilleRisLambers, & Levine, 2007) that combine neutral zero-sum demographic drift with weak stabilizing and equalizing coexistence mechanisms.

What does the "zero-sum" assumption mean in Hubbell's Zero-Sum Multinomial (ZSM)?

The ecological zero-sum assumption states that a local guild of trophically similar organisms (such as trees in a closed forest canopy) saturates all available limiting resources and space, so the total number of individuals in the local community remains constant at J_L = ∑ n_i. Because total community size is fixed, every time one individual dies, one new individual is recruited to fill the vacancy—meaning that one species' population gain (+1) must always be balanced by another species' loss (−1), so that ∑ Δn_i = 0.

How do the two parameters of the ZSM distribution (θ and m) control the shape of a community's species abundance curve?

The fundamental biodiversity number (θ = 2 J_M ν) governs the regional metacommunity richness and sets the overall height and rare-species tail of the abundance distribution (higher θ produces a more hyperdiverse regional species pool). The immigration probability (m ∈ (0, 1]), often expressed via the coupling parameter γ = m(J_L − 1)/(1 − m), controls the degree of local dispersal limitation: when m = 1 (no dispersal limitation), the local ZSM equals Fisher's logseries with maximum singletons; as m decreases toward 0 (strong dispersal limitation), rare immigrants frequently go locally extinct before establishing, depleting singletons and shifting the Preston octave curve into a left-skewed hump with an interior mode.

Why does the ZSM outperform Preston's lognormal distribution in tropical rainforest tree plots?

Preston's lognormal distribution assumes that logarithmic species abundances follow a symmetric Gaussian bell curve on log₂ abundance octaves. However, real census data from large tropical forest plots (such as the 50-hectare Barro Colorado Island plot in Panama and Pasoh plot in Malaysia) consistently exhibit a pronounced negative (left) skewness—containing substantially more rare species in the low-abundance octaves than a symmetric Gaussian curve predicts, yet fewer singletons than Fisher's logseries predicts. By coupling local zero-sum birth–death drift (J_L) to metacommunity immigration (m ≈ 0.08–0.15), the ZSM naturally generates this left-skewed octave profile from first principles without adding arbitrary curve-fitting parameters.

1. McKane, A., Alonso, D., & Solé, R. V. (2000). "Mean-field stochastic theory for species-rich assembled communities." Physical Review E, 62(6), 8466–8484. https://doi.org/10.1103/PhysRevE.62.8466
2. Hubbell, S. P. (2001). The Unified Neutral Theory of Biodiversity and Biogeography (Monographs in Population Biology, Vol. 32). Princeton, NJ: Princeton University Press. https://press.princeton.edu/books/paperback/9780691021287/the-unified-neutral-theory-of-biodiversity-and-biogeography
3. Volkov, I., Banavar, J. R., Hubbell, S. P., & Maritan, A. (2003). "Neutral theory and relative species abundance in ecology." Nature, 424(6952), 1035–1037. https://doi.org/10.1038/nature01883
4. Etienne, R. S. (2005). "A new sampling formula for neutral biodiversity." Ecology Letters, 8(3), 253–260. https://doi.org/10.1111/j.1461-0248.2004.00717.x
5. Alonso, D., Etienne, R. S., & McKane, A. J. (2006). "The merits of neutral theory." Trends in Ecology & Evolution, 21(8), 451–457. https://doi.org/10.1016/j.tree.2006.03.019