Binomial Maps in Finite-Population Dynamics

• Binomial maps are structured transformations that reinterpret deterministic recurrences as state-dependent binomial or Poisson sampling in finite-population models.
• They bridge discrete deterministic dynamics and stochastic agent-based models by incorporating demographic noise, extinction risks, and the Jensen gap in nonlinear updates.
• In combinatorics and integrable systems, binomial maps underlie subsequence count invariants and yield Yang–Baxter maps via first-degree Lax matrices.

Searching arXiv for papers directly relevant to “Binomial Maps” and closely related uses of the term. Binomial maps are stochastic finite-population counterparts of deterministic discrete-time population maps in which the deterministic update is interpreted as a conditional mean and the realized next state is sampled from an integer-valued distribution, binomial in the bounded case and Poisson in the unbounded limit (Shekatkar, 16 Aug 2025). The same expression, or closely related “binomial map” language, also appears in other mathematical settings: in combinatorics on words it refers to the subsequence-count map $u\mapsto \bigl(\binom{u}{v}\bigr)_{|v|\le k}$ underlying $k$-binomial equivalence and $k$-binomial complexity (Lejeune et al., 2018), while in integrable systems it occurs through Yang–Baxter maps built from binomial Lax matrices $L(\zeta)=X-\zeta A$ (Kouloukas et al., 2011). The term is therefore context-dependent, but in each usage it denotes a structured transformation governed by binomial sampling, binomial subsequence counts, or first-degree “binomial” matrix polynomials.

1. Finite-population formulation in discrete-time dynamics

In population dynamics, the starting point is a deterministic map

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$

where $\mathbf{X}(t)$ is the state vector at time $t$, each component represents an average population or abundance, and $\Theta$ denotes model parameters. Such maps include the Logistic, Ricker, Beverton–Holt, and Nicholson–Bailey models. Their deterministic character is defensible in an infinite-population or continuum limit, because fluctuations are negligible relative to the mean, but it is ill-suited to finite populations, where counts are integer-valued, fluctuations can be large at small population size, and extinction can occur purely from chance (Shekatkar, 16 Aug 2025).

The construction of a Binomial map introduces an availability vector

$$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$

whose components $A_{i,t}$ represent maximal available counts, slots, or agents for component $k$0. The stochastic state $k$1 has nonnegative integer components, and the key modeling assumption is that $k$2 determines only the conditional mean of $k$3, not the next state exactly. For component $k$4,

$k$5

and

$k$6

provided

$k$7

This yields a state-dependent transition kernel on nonnegative integers. If components are sampled independently conditional on the current state, the full transition kernel factors across components. The mean is fixed by the original deterministic map, while the variance emerges from finite sampling: $k$8 The corresponding relative fluctuation size is

$k$9

Thus the noise amplitude is not imposed exogenously; it is generated by Bernoulli or binomial sampling and decreases relative to the mean as population scale grows.

A central conceptual distinction is between demographic and environmental stochasticity. Demographic stochasticity is internal noise arising from the discreteness and chance behavior of individuals; environmental stochasticity is external noise produced by weather variation, resource shocks, or habitat disturbances. The finite-population Binomial-map construction is designed to model the former, not merely to add a random perturbation to a deterministic recurrence. Additive or externally imposed noise can model environmental variation, but it does not automatically preserve integer states, state-dependent variance, or the correct large-population scaling for demographic noise.

2. Agent-based interpretation and canonical examples

A major feature of the Binomial-map formalism is that it makes the connection to agent-based models explicit. If component $k$0 has $k$1 available individuals or sites and each becomes a success at time $k$2 with the same probability $k$3, then

$k$4

Consistency with the deterministic map requires

$k$5

The Binomial map is therefore a coarse-grained agent-based model with exchangeable Bernoulli trials, identical individuals within each component, conditional independence given the current state, and a deterministic update law specifying the conditional mean (Shekatkar, 16 Aug 2025).

The bounded prototype in the paper is the Logistic map

$k$6

where $k$7 is the carrying capacity. Since the population cannot exceed $k$8, one takes $k$9, giving

$L(\zeta)=X-\zeta A$0

and

$L(\zeta)=X-\zeta A$1

In fraction form $L(\zeta)=X-\zeta A$2, the mean-field update is $L(\zeta)=X-\zeta A$3 for large $L(\zeta)=X-\zeta A$4.

The unbounded prototype is the Ricker map

$L(\zeta)=X-\zeta A$5

Here there is no finite upper bound, so the availability is effectively infinite. Taking the limit $L(\zeta)=X-\zeta A$6 and using the standard Poisson limit of the binomial,

$L(\zeta)=X-\zeta A$7

one obtains the Poisson-map counterpart

$L(\zeta)=X-\zeta A$8

$L(\zeta)=X-\zeta A$9

These two examples delimit the bounded and unbounded regimes. In bounded state spaces, binomial sampling is natural because there is a finite maximum number of possible successes at each step. In unbounded state spaces, the same construction passes to a Poisson law. This suggests a unified finite-population interpretation of familiar iterated maps without leaving the discrete-time setting.

3. Deterministic recovery and the Jensen-gap obstruction

Binomial maps recover deterministic behavior only asymptotically. Relative fluctuations vanish when expected population and capacity become large, so trajectories concentrate around the deterministic update in a law-of-large-numbers sense. For the Logistic map this occurs as $\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$0; for the Ricker map it occurs as the population scale set by $\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$1 grows (Shekatkar, 16 Aug 2025).

At finite size, however, the deterministic map is not generally the same as the mean stochastic trajectory. If

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$2

and the deterministic and stochastic systems start from the same initial condition $\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$3, then at one time step

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$4

At the next step,

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$5

For nonlinear $\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$6, these are not equal in general. The discrepancy is the Jensen gap

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$7

For analytic $\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$8,

$\mathbf{X}(t+1)=\mathbf{f}(\mathbf{X}(t),\Theta),$9

so

$\mathbf{X}(t)$0

A sufficient condition for $\mathbf{X}(t)$1 in the infinite-agent limit is that

$\mathbf{X}(t)$2

All linear maps satisfy this trivially, so they are equivalent to their deterministic counterparts.

The Logistic and Ricker maps satisfy the sufficient condition in their respective large-capacity limits. For

$\mathbf{X}(t)$3

one has

$\mathbf{X}(t)$4

hence $\mathbf{X}(t)$5 as $\mathbf{X}(t)$6. For the Ricker map,

$\mathbf{X}(t)$7

and for $\mathbf{X}(t)$8,

$\mathbf{X}(t)$9

so $t$0 as $t$1 for all $t$2. The multidimensional extension replaces ordinary derivatives by mixed partial derivatives of all orders.

This non-equivalence is a central theoretical correction. It rules out the naive identification of “deterministic map = mean stochastic trajectory” except at one step or under additional conditions.

4. Extinction, quasi-stationarity, and finite-size effects

Because Binomial maps are integer-valued and state dependent, they admit extinction without external forcing. In the Binomial Logistic map, $t$3 is absorbing: once extinct, the population remains extinct. The Poisson Ricker map can also hit zero, so extinction is possible there as well (Shekatkar, 16 Aug 2025).

The extinction mechanism is demographic. Extinction probability is larger when the expected next population is small, and finite populations with smaller carrying capacity are therefore at greater risk. In the Logistic example, the paper reports numerically that small $t$4 produces substantial trajectory fluctuations, while large $t$5 suppresses them and the bifurcation diagram approaches that of the deterministic logistic map. The same qualitative convergence holds for the Ricker example as $t$6 increases, but unlike the deterministic recurrence, the stochastic model still allows extinction.

The paper also studies extinction times $t$7 for the Binomial Logistic map at $t$8, a chaotic deterministic regime, and reports that the extinction-time distributions are approximately discrete-exponential, except at $t$9 and $\Theta$0. This positions Binomial maps as discrete-time analogues of finite-population continuous-time simulations based on Gillespie-type methods: many stochastic trajectories can be generated and the full extinction-time distribution estimated.

The broader program is explicitly finite-population analysis within map-based models. The framework is presented as opening the door to rigorous study of extinction risk, quasi-stationarity, and noise-induced transitions. At the same time, several foundational dynamical-systems questions remain open, including how to define stability in the stochastic setting, how to characterize sensitive dependence on initial conditions, how to analyze bifurcations under stochasticity, and how to describe long-term behavior in the presence of fluctuations.

5. Binomial maps on words and $\Theta$1-binomial complexity

In combinatorics on words, “binomial map” language refers to subsequence-count invariants. For finite words $\Theta$2, the binomial coefficient of words

$\Theta$3

counts the number of occurrences of $\Theta$4 as a subsequence of $\Theta$5. Two words are $\Theta$6-binomially equivalent if

$\Theta$7

For an infinite word $\Theta$8, the $\Theta$9-binomial complexity is

$$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$0

and the associated binomial map may be written as

$$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$1

Then $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$2 exactly when $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$3 (Lejeune et al., 2018).

The classical Thue–Morse word provides the canonical explicit example. For every positive integer $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$4,

$$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$5

while for all $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$6,

$$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$7

The threshold is exactly $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$8. Below it, no two distinct factors of length $$\mathbf{A}_t=\boldsymbol{\phi}(\mathbf{X}(t),\Theta),$$9 are $A_{i,t}$0-binomially equivalent; from $A_{i,t}$1 onward the complexity takes only two values depending on $A_{i,t}$2. For $A_{i,t}$3, this gives

$A_{i,t}$4

and for all $A_{i,t}$5,

$A_{i,t}$6

The structural explanation is boundary alignment relative to the Thue–Morse block decomposition. For factors $A_{i,t}$7 of length $A_{i,t}$8,

$A_{i,t}$9

where $k$00 and $k$01 are order-$k$02 factorization pairs and $k$03 is the boundary-pair equivalence relation introduced for the proof. In this regime, the entire bounded-length subsequence-count vector is determined by order-$k$04 type. This suggests that the binomial map $k$05 becomes a coarse boundary invariant once factor length exceeds the intrinsic substitution scale.

6. Morphisms, boundedness transfer, and generalized Thue–Morse words

A second line of work studies how morphisms act on binomial complexity. For a morphism $k$06, the relevant class is that of Parikh-collinear morphisms, equivalently morphisms whose adjacency matrix has rank $k$07. These are exactly the morphisms that raise boundedness one level in the binomial hierarchy: $k$08 Equivalently, for all $k$09,

$k$10

A direct corollary is that fixed points of Parikh-collinear morphisms have bounded $k$11-binomial complexity for every $k$12 (Rigo et al., 2022).

The Thue–Morse morphism

$k$13

is Parikh-constant, hence Parikh-collinear. It satisfies

$k$14

which explains why powers of $k$15 can wash out lower-order binomial distinctions while preserving higher-order ones. The same paper gives a new characterization of Sturmian words: if for some fixed $k$16,

$k$17

then $k$18 is Sturmian.

For generalized Thue–Morse words $k$19 over the alphabet $k$20, generated by the $k$21-uniform substitution

$k$22

the $k$23-binomial complexity has now been computed explicitly for every $k$24 and $k$25. The shortest pair of distinct $k$26-binomially equivalent factors has length $k$27, so

$k$28

For $k$29, if

$k$30

then

$k$31

The paper states that the periodic part begins at $k$32, with period $k$33 (Golafshan et al., 2024).

The structural criterion parallels the binary case. Two factors $k$34 of $k$35 are $k$36-binomially equivalent if and only if there exist $k$37-factorizations

$k$38

such that

$k$39

This reduces $k$40-binomial equivalence to identical short boundaries and abelian equivalence of desubstituted cores. A plausible implication is that, in this literature, “binomial map” is best understood not as a single named object but as a family of subsequence-count maps and morphic transformations governing a hierarchy of equivalence relations.

7. Binomial Lax matrices and Yang–Baxter maps

In integrable-systems literature, the phrase occurs in a different and more specialized sense. “Binomial” refers not to sampling or subsequence counts but to first-degree matrix polynomials

$k$41

The paper “Poisson Yang-Baxter maps with binomial Lax matrices” develops parametric Yang–Baxter maps from such Lax matrices by imposing the refactorization identity

$k$42

together with preservation of the Casimirs of the Sklyanin Poisson bracket (Kouloukas et al., 2011).

In the $k$43 case, this yields an explicit parametric Yang–Baxter map on matrix space; restriction to symplectic leaves produces lower-dimensional symplectic Yang–Baxter maps with strong Lax matrices. The construction extends to $k$44 first-degree matrix polynomials and is worked out in detail for a $k$45 family giving quadrirational symplectic Yang–Baxter maps on $k$46. The paper also connects specializations and degenerations of the construction to the Adler–Yamilov, Boussinesq, and Goncharenko–Veselov maps.

This usage is terminologically narrower than the population-dynamical one. The paper explicitly notes that the phrase “binomial maps” is not really a technical class name for the maps themselves; rather, the maps are Yang–Baxter maps derived from binomial, that is first-degree, Lax matrices. The common feature across usages is therefore structural sparsity: a deterministic map becomes a binomially sampled stochastic update, a word becomes a vector of binomial subsequence counts, or a Lax representation is built from a two-term matrix polynomial.