FileURL: https://81018.com/constraint-classes-ongoing-synthesis/ Also: https://81018.com/constraint-links/
On-going discovery process
Constraint Classes in Discrete Geometric Growth Models
A Mathematical Taxonomy of Constrained Discrete Growth
Note: Synthetic version was “cleaned” by Anthropic’s Claude — mathematics only, no physical interpretation.
Abstract
6 August 2026. Discrete recurrence relations provide a natural mathematical framework for investigating systems that evolve through successive states. While unconstrained exponential growth is increasingly well-understood, relatively little attention has been given to the systematic classification of growth processes modified by general constraint functions. This paper develops a taxonomy of constrained discrete growth models based on the recurrence relation N_{n+1} = 2N_n − ε(N_n), where ε(N_n) is a non-negative constraint function. Rather than proposing a specific physical theory, the objective is to classify broad families of constraint functions and examine the characteristic expansion regimes they generate. Representative classes — including power-law, logarithmic, threshold, delayed, and stochastic constraints — are introduced and their qualitative asymptotic behaviors compared. The resulting framework provides a mathematical basis for studying constrained discrete growth independently of any particular physical interpretation.
This paper proceeds as follows: Section 2 introduces the general recurrence framework. Section 3 defines six constraint classes, while Section 4 classifies their asymptotic behaviors into seven regimes. Sections 5–6 illustrate these with examples and relate them to existing work. Section 7 discusses open mathematical questions, and Section 8 concludes.
1. Introduction
1.1 Why classify?
Classification has often preceded explanation in the history of science. The periodic table organized the chemical elements before quantum mechanics explained atomic structure. Biological taxonomy preceded evolutionary theory. Stellar classification provided an observational framework before modern stellar evolution was understood. In each case, organizing related phenomena revealed patterns that later became the foundation of deeper theoretical understanding.
The present work adopts the same philosophy. Rather than beginning with a comprehensive physical theory, it asks whether constrained discrete growth processes themselves admit a useful mathematical classification. If distinct families of constraint functions produce distinct classes of large-scale behavior, such a taxonomy may provide a common language for comparing models developed in different areas of mathematics and physics.
1.2 Why recurrence relations?
Many physical theories are formulated through differential equations that assume continuity. If, however, the underlying evolution of a system is fundamentally discrete, recurrence relations become the natural mathematical language. They describe evolution one state at a time and allow local rules to generate global behavior through iteration.
Recurrence relations already occupy central roles in dynamical systems, branching processes, computer science, population biology, combinatorics, and numerical analysis. Here they are used as a minimal framework for studying discrete growth under constraint.
1.3 Why constraints?
Ideal or pure exponential growth, N_{n+1} = 2N_n, is mathematically simple but physically exceptional. Most natural systems experience limitations arising from geometry, interactions, finite resources, boundary conditions, or structural organization. Such effects may be represented abstractly through a constraint function,
N_{n+1} = 2N_n − ε(N_n).
Rather than proposing a single form for ε, this paper investigates families of constraint functions and the dynamical behaviors they produce.
1.4 Scope
The purpose of this paper is deliberately modest. It does not propose a physical theory of any specific system. It does not derive equations of motion for any concrete application. Instead, it asks a mathematical question:
What classes of large-scale behavior emerge from different classes of constraints acting on a simple discrete growth process?
The answer is a taxonomy.
2. General Framework
We consider the recurrence relation
N_{n+1} = 2N_n − ε(N_n),
where N_n denotes the size of the system after the n-th iteration and ε(N_n) is a non-negative constraint function, with:
- N_n > 0
- ε(N_n) ≥ 0
- growth remains monotonic unless otherwise stated.
A constraint function ε is called admissible if it satisfies ε(N) ≥ 0 for all N > 0, is locally bounded, and does not dominate the growth term — that is, ε(N) < 2N for all N in the relevant range. The first condition ensures monotonicity is maintained except where explicitly broken. The second ensures the recurrence remains well-defined. The third ensures the system does not collapse to zero in finite steps. Beyond these minimal conditions, the constraint function may take any functional form, and the variety of admissible forms is the subject of Section 3.
3. Constraint Classes
The following six classes are defined by the functional form of ε(N) and distinguished by the asymptotic behavior they produce.
Class I — No Constraint
Definition: ε(N) = 0 for all N.
Recurrence: N_{n+1} = 2N_n, with solution N_n = N₀ · 2ⁿ.
Asymptotic behavior: Pure exponential growth. The system doubles at every step without modification.
Dynamical character: This is the unconstrained baseline against which all other classes are measured. It is dynamically simple, analytically solvable, and physically exceptional — no natural system sustains ideal exponential growth indefinitely. Its inclusion in the taxonomy is essential precisely because it defines the reference regime from which all constrained classes deviate.
Literature connection: Unconstrained doubling recurrences appear throughout combinatorics, computer science, and branching process theory.
Class II — Power-Law Constraints
Definition: ε(N) = κNᵅ, with κ > 0 and 0 < α < 1.
Recurrence: N_{n+1} = 2N_n − κNₙᵅ.
Asymptotic behavior: For small κ, the leading behavior at large n remains near-exponential. The constraint produces sub-extensive corrections: N_n ~ 2ⁿ · (1 − κ · C · 2^((α−1)n)) for some constant C depending on α and the initial condition. Because α < 1, the correction term decays relative to the dominant exponential, and the system asymptotically approaches Class I behavior while remaining permanently suppressed below it.
At α → 1 the constraint becomes linear and may be absorbed into a rescaled growth rate: N_{n+1} = (2 − κ)N_n, giving N_n = N₀ · (2 − κ)ⁿ. This is still exponential but with a reduced base.
At α = 0 the constraint becomes constant: ε = κ, a fixed deduction at every step. This boundary case sits between Class I and Class II.
Dynamical character: Near-exponential growth with permanent sub-leading suppression. The system expands without bound but always below the Class I trajectory. The gap between constrained and unconstrained trajectories widens in absolute terms but shrinks in relative terms.
Literature connection: Power-law constraints arise in systems with geometric frustration, where the number of defect sites scales as a fractional power of the system size. They also appear in population dynamics under Allee effects and in network growth with preferential attachment.
Class III — Logarithmic Constraints
Definition: ε(N) = κ log N, with κ > 0 (any fixed base).
Recurrence: N_{n+1} = 2N_n − κ log N_n.
Asymptotic behavior: Because ε(N)/N = κ log N/N → 0 as N → ∞, the constraint becomes negligible relative to the doubling term. The trajectory remains asymptotically exponential:
N_n = a_∞ · 2ⁿ + κ(ln 2) · n + O(1),
where a_∞ > 0 depends on κ and the initial condition. The system therefore belongs to the near-exponential regime (Regime II). The absolute gap relative to the unconstrained trajectory grows, but the relative effect of the constraint vanishes.
Dynamical character: Permanent but asymptotically weak suppression. The instantaneous ratio N_{n+1}/N_n → 2 from below; the deviation from 2 decays as O(n/2ⁿ).
Interpretation: Logarithmic constraints form a useful weak-correction class for systems in which the constraint grows slowly with system size. They never produce a pure power-law (polynomial) regime.
Literature connection: Logarithmic constraints appear in information-theoretic contexts (entropy scaling), statistical mechanics near critical points, algorithm analysis, and models in which the number of distinct edge types or information-processing capacity grows logarithmically with system size.
Class IV — Threshold Constraints
Definition: ε(N) is piecewise, with qualitatively different behavior above and below a threshold value N*:
ε(N) = ε₁(N) for N < N*
ε(N) = ε₂(N) for N ≥ N*
where ε₁ and ε₂ belong to different constraint classes or have qualitatively different magnitudes.
Recurrence: N_{n+1} = 2N_n − ε(N_n), with ε switching form at N = N*.
Asymptotic behavior: Threshold constraints produce systems with distinct dynamical phases separated by a transition at N*. Below the threshold the system evolves according to one regime; above it according to another. If ε₂ is strong enough — specifically if ε₂(N) ≥ N for all N ≥ N* — the system saturates at or near N* rather than continuing to grow. If ε₂ is weaker, the system continues to grow but with a different rate characteristic than below the threshold.
The transition at N* may be sharp or gradual depending on the continuity of ε at N*. A discontinuous jump in ε at N* produces an abrupt phase transition in the growth trajectory; a smooth transition produces a gradual crossover.
Dynamical character: Multi-phase growth with at least one transition. The system may exhibit qualitatively different behavior at different scales — for example, near-linear growth below the threshold followed by exponential growth above it, or exponential growth followed by saturation. This class is the natural mathematical language for systems that undergo symmetry breaking or phase transitions as they evolve.
Literature connection: Threshold constraints are ubiquitous in physics and biology. They appear in percolation theory, in models of phase transitions, in neural network activation functions, and in epidemiological models with herd immunity thresholds.
Class V — Delayed Constraints
Definition: ε depends on an earlier state rather than the current state:
ε = ε(N_{n−τ})
where τ ≥ 1 is the delay in steps.
Recurrence: N_{n+1} = 2N_n − ε(N_{n−τ}).
Asymptotic behavior: Delayed constraints introduce memory into the system. The current growth rate depends on the state τ steps ago, creating a feedback loop between past and present. For τ = 1 the system has one step of memory; for larger τ the memory extends further into the past.
The asymptotic behavior of Class V systems depends sensitively on the form of ε and the delay τ. For linear ε the system admits analysis via characteristic polynomials of degree τ + 1, which may have complex roots, producing oscillatory behavior. For nonlinear ε, more complex dynamics including quasi-periodicity and chaos are possible. In general, increasing τ tends to destabilize otherwise stable growth trajectories, and the boundary between stable near-exponential growth and oscillatory or chaotic behavior depends on the product κτ for power-law ε.
Dynamical character: Potentially oscillatory, quasiperiodic, or chaotic, depending on delay and constraint strength. Class V systems are the most dynamically rich in the taxonomy. They also have the most complex analytical structure, as the recurrence relation becomes a delay-difference equation requiring specification of τ initial conditions rather than one.
Literature connection: Delay-difference equations appear throughout population biology (predator-prey models with delayed reproduction), economics (business cycle models), control theory, and epidemiology. In mathematical physics, delayed constraints arise when a system’s current behavior is regulated by information that takes a finite time to propagate through the system — a natural feature of any framework with a finite signal speed.
Class VI — Stochastic Constraints
Definition: ε(N_n) = ε̄(N_n) + σ · ξ_n
where ε̄(N_n) is a deterministic constraint belonging to one of Classes I–V, σ ≥ 0 is a noise amplitude, and ξ_n is a random variable with zero mean and unit variance — for example, ξ_n ~ N(0,1) or ξ_n uniform on [−1, 1].
Recurrence: N_{n+1} = 2N_n − ε̄(N_n) − σ · ξ_n.
Asymptotic behavior: The behavior of Class VI systems is characterized by a distribution of trajectories rather than a single trajectory. The mean trajectory ⟨N_n⟩ follows approximately the deterministic Class I–V behavior corresponding to ε̄, while the variance of N_n grows with n at a rate depending on σ and the amplification properties of the doubling map.
For the pure doubling map (ε̄ = 0), perturbations are amplified exponentially: a noise contribution at step k is doubled n − k times by step n, so the variance of N_n grows as σ² · (4ⁿ − 1)/3, exponentially faster than the mean trajectory. This means that even small noise amplitudes eventually produce trajectories that deviate substantially from the mean — a fundamental instability that must be accounted for in any stochastic extension of the framework.
For constrained ε̄, the amplification rate is reduced, and the variance may grow more slowly depending on the constraint class.
Dynamical character: Ensemble of trajectories with statistical distribution. Individual realizations may differ substantially from the mean, and the distribution of N_n may be asymmetric or heavy-tailed depending on the noise model. Class VI is the appropriate framework for systems in which the constraint itself is subject to uncertainty — either from measurement limitations, from genuine indeterminacy, or from sensitivity to initial conditions.
Literature connection: Stochastic recurrence relations are studied extensively in probability theory, mathematical finance, and statistical physics. They appear in models of genetic drift, random walks in random environments, and multiplicative noise processes.
4. Expansion Taxonomy
The six constraint classes defined in Section 3 produce seven qualitatively distinct asymptotic regimes. Rather than classifying the constraint functions themselves, this section classifies the outcomes — the large-n behavior of the trajectory N_n. The result is an expansion taxonomy that may be applied to any constrained discrete growth model regardless of its physical interpretation.
4.1 Exponential regime
The exponential regime is defined by N_{n+1}/N_n = 2 exactly for all n. It is produced only by Class I (ε = 0) and represents the mathematical ideal against which all constrained regimes are measured. No constrained system with ε > 0 belongs to this regime, though several approach it asymptotically.
The exponential regime is the only regime in which the doubling time is exactly constant — one step — and in which the trajectory is fully determined by a single parameter N₀.
4.2 Near-exponential regime
The near-exponential regime is defined by N_{n+1}/N_n → 2 as n → ∞, with the ratio approaching 2 from below but never reaching it in finite steps. This regime is produced by Class II with 0 < α < 1 (sub-linear power-law constraints) and by Class III (logarithmic constraints). In both cases, the constraint becomes negligible relative to the doubling term at large N.
The near-exponential regime is the most physically relevant regime for models of sustained large-scale growth in general, because it permits sustained growth while incorporating a persistent but asymptotically weak constraint. Systems in this regime eventually become indistinguishable from pure exponential growth on the largest scales, yet retain measurable sub-leading corrections at intermediate scales — corrections that are, in principle, the observable signature of whatever constraint is at work in a given application.
4.3 Reduced exponential regime
The reduced exponential regime is defined by N_{n+1}/N_n → r for some constant 1 < r < 2. It is produced by Class II at α = 1, where the linear constraint ε = κN reduces the effective growth rate from 2 to 2 − κ, and by Class IV systems in which the above-threshold constraint is linear and sustained.
This regime is exponential in character but with a reduced base. The trajectory is still unbounded but grows more slowly than the Class I baseline by a constant multiplicative factor at each step. Systems in this regime are analytically tractable — the solution N_n = N₀ · (2 − κ)ⁿ is closed form — and provide a useful intermediate case between full exponential growth and the polynomial regime.
4.4 Polynomial regime
The polynomial regime is defined by N_n ~ C·nᵖ for some p > 1.
Pure logarithmic constraints (Class III) never produce this regime; they remain near-exponential for every fixed κ > 0. A genuine polynomial regime requires a stronger constraint that continues to suppress the leading multiplier below 2 in a sustained way — for example forms such as ε(N) = κN/log N, or certain threshold constructions that keep the effective growth rate near 1.
On a log-log plot such trajectories appear linear, providing a simple diagnostic. Systems in the polynomial regime are unbounded but grow dramatically more slowly than any exponential process.
4.5 Saturating regime
The saturating regime is defined by N_n → N* as n → ∞ for some finite fixed point N*. It is produced by Class IV systems in which the above-threshold constraint is strong enough to balance the doubling term — specifically when ε₂(N*) = N* exactly, so that N_{n+1} = 2N* − N* = N*.
The fixed point N* is determined by the condition ε(N*) = N*, which has at least one solution whenever ε is continuous and ε(N)/N → ∞ as N → ∞. The stability of the fixed point depends on the derivative ε'(N*): the fixed point is asymptotically stable when 1 < ε'(N*) < 3 and unstable otherwise.
Saturating systems represent growth processes that are permanently bounded by their constraint — the biological carrying-capacity model is the canonical example. Saturation provides a useful limiting case for calibrating the strength of a constraint function in any model where unbounded growth is the expected or observed behavior: a constraint that produces saturation at realistic parameter values is, by construction, too strong to be consistent with sustained growth in such a model.
4.6 Oscillatory regime
The oscillatory regime is defined by trajectories that alternate above and below a mean value rather than converging monotonically. It is produced primarily by Class V systems with delay τ ≥ 1, where the feedback between past and present states generates overshoot and correction cycles.
The oscillatory regime admits two sub-cases. In the bounded oscillatory sub-case, the amplitude of oscillations remains finite as n → ∞, and the trajectory remains near its mean. In the growing oscillatory sub-case, the amplitude increases with n, and the trajectory eventually diverges. The boundary between these sub-cases is determined by the stability of the characteristic polynomial of the delay-difference equation.
Its inclusion in the taxonomy is motivated by completeness: any system whose constraint mechanism has a genuine time lag — whether from finite signal propagation speed, processing delay, or a lagged feedback loop — is a candidate for this regime.
4.7 Stochastic regime
The stochastic regime is not a single trajectory but a distribution of trajectories. It is produced by Class VI systems and is characterized by the mean trajectory ⟨N_n⟩ and the variance Var(N_n), both of which evolve with n.
For the pure doubling map with additive noise (ε̄ = 0), the variance grows as σ² · (4ⁿ − 1)/3 — exponentially faster than the mean — meaning that individual trajectories diverge from the ensemble mean at an exponential rate. This is a fundamental property of multiplicative noise in doubling systems and implies that even small perturbations become significant over many steps.
For constrained ε̄, the variance growth rate is reduced, and the distribution of N_n may become approximately Gaussian at large n by central limit arguments if the noise contributions at different steps are sufficiently independent. Assessing whether any deterministic result derived from a Class I–V model survives moderate stochastic perturbation is, in general, exactly the kind of question Class VI provides the tools to investigate.
| Regime | Asymptotic behaviour of N_n | Growth rate | Produced by | Key signature |
|---|---|---|---|---|
| I. Exponential | N_n ~ N₀·2ⁿ | Doubles every step | Class I (ε=0) | Constant ratio N_{n+1}/N_n=2 |
| II. Near-exponential | N_n ~ C·2ⁿ·f(n), f→1 | Asymptotically doubling | Class II (0<α<1), Class III | Ratio → 2 from below |
| III. Reduced exponential | N_n ~ N₀·rⁿ, 1<r<2 | Exponential, reduced base | Class II (α=1), some Class IV | Constant ratio r<2 |
| IV. Polynomial | N_n ~ C·nᵖ, p>1 | Power-law | Stronger-than-logarithmic forms | Log-log linear growth |
| V. Saturating | N_n → N* | Zero asymptotically | Class IV (strong ε₂≥N) | Trajectory flattens to fixed point |
| VI. Oscillatory | Oscillates around a mean | Bounded or growing oscillation | Class V (delay ≥1) | Alternating over/undershoot |
| VII. Stochastic | Distribution over trajectories | Depends on ε̄ | Class VI | Ensemble spread grows with n |
5. Representative Examples
This section illustrates each expansion regime with a representative numerical simulation. Each figure shows N_n versus n on both linear and log scales. The grey dashed line in each figure is the Class I exponential baseline (ε = 0), providing a visual reference against which the effect of each constraint class can be measured. All simulations begin at N₀ = 1 and run for 60 iterations.
The simulations in this section are not intended as physical predictions. They are canonical exemplars — each chosen to lie cleanly within one regime of the taxonomy. Their purpose is to make the diagnostic criteria in Table 1 visually concrete. A reader encountering a new constrained growth process can compare its trajectory to these exemplars and, using the diagnostics, classify it without needing to derive its asymptotic form analytically. Parameter choices are illustrative rather than physically calibrated; each is accompanied by a rationale and selected to keep the constraint admissible (ε(N) < 2N for all n ≤ 60) while producing a clearly visible deviation from the baseline within 60 steps.
5.1 Regime I — Exponential
With no constraint (ε = 0), the recurrence N_{n+1} = 2N_n produces pure exponential growth. The trajectory doubles at every step and appears as a straight line on the log scale. This is the reference baseline.
Constraint: ε(N) = 0 for all N.
Recurrence: N_{n+1} = 2N_n, with solution N_n = N₀ · 2ⁿ.
Rationale: This is the unconstrained baseline. No parameters are required. Its inclusion establishes the reference trajectory against which all constrained regimes are compared.
Diagnostic: Plot the ratio R_n = N_{n+1}/N_n versus n. In Regime I, R_n = 2 exactly for all n. On a semi-log plot (log N_n vs. n), the trajectory is a perfect straight line with slope log 2.
5.2 Regime II — Near-Exponential
With a power-law constraint ε = 0.05 · N^0.5, the trajectory grows near-exponentially but remains permanently below the Class I baseline. On the log scale the separation between the constrained trajectory and the baseline is visible but small and decreasing in relative terms.
Constraint: ε(N) = κNᵅ, with κ = 0.05 and α = 0.5.
Recurrence: N_{n+1} = 2N_n − 0.05√N_n.
Rationale: This choice places the constraint in the interior of Class II (0 < α < 1). The value κ = 0.05 is small enough that the constraint remains weak relative to the doubling term (0.05√N ≪ 2N for N ≫ 1), ensuring near-exponential growth, but large enough that the cumulative suppression over 60 steps produces a visually perceptible gap from the baseline on the semi-log plot.
Diagnostic: Compute the ratio R_n = N_{n+1}/N_n. In Regime II, R_n < 2 for all finite n, but R_n → 2 monotonically from below as n → ∞. For Class II with 0 < α < 1, the approach is 2 − R_n ~ C·n^{α−1}, a power-law decay. This distinguishes Regime II from Regime III (where R_n → r < 2, a constant) and Regime I (where R_n = 2 exactly).
5.3 Regime III — Reduced Exponential
Constraint: ε(N) = κN with κ = 0.30.
Recurrence: N_{n+1} = (2 − 0.30)N_n = 1.70·N_n.
Rationale: A linear constraint (α = 1) simply rescales the growth rate. The trajectory remains purely exponential but with a reduced base. This is the analytically simplest intermediate case between full exponential growth and more strongly suppressed regimes.
Diagnostic: The ratio R_n = N_{n+1}/N_n is exactly constant and equal to 1.70 < 2 for all n. On a semi-log plot the trajectory is a straight line with slope log 1.70.
5.4 Regime IV — Polynomial
Pure logarithmic constraints do not produce polynomial growth. A representative stronger form that can generate this regime is ε(N) = κN/log N (with κ > 0 chosen so that the effective multiplier stays below 2). On a log-log plot the resulting trajectory appears approximately linear. Detailed numerical exemplars of such stronger constraints are left for a future revision; the present section simply records that the polynomial regime exists and is distinct from every pure-logarithmic trajectory.
5.5 Regime V — Saturating
Constraint: ε(N) = 0 for N < 10⁶; ε(N) = 1.5N for N ≥ 10⁶.
Recurrence: N_{n+1} = 2N_n below threshold; N_{n+1} = 0.5N_n above threshold.
Rationale: The threshold N* = 10⁶ is reached near step 20 under pure doubling from N₀ = 1, producing a clearly visible two-phase trajectory. The above-threshold coefficient 1.5 > 1 ensures the fixed-point condition ε(N*) = N* is satisfied and the trajectory saturates rather than oscillating.
Diagnostic: The ratio R_n = N_{n+1}/N_n transitions sharply from 2 (below threshold) to approximately 0.5 (immediately above threshold), then converges to 1 as n → ∞. This three-phase ratio behavior is unique to the saturating regime.
5.6 Regime VI — Oscillatory
Constraint: Delayed constraint ε(N_n) = κ·N_{n−τ}, with κ = 0.95 and delay τ = 2.
Recurrence: N_{n+1} = 2N_n − 0.95·N_{n−2}.
Rationale: The delay τ = 2 introduces memory of the state two steps ago, creating a feedback loop between past and present. The coefficient κ = 0.95 is chosen to be large enough to produce visible oscillations but below the threshold for immediate instability. For κ = 0.95, the characteristic polynomial has complex roots with magnitude slightly greater than 1, producing growing oscillations — a hallmark of delayed feedback in discrete systems. This regime illustrates the dynamical richness that memory effects can introduce.
5.7 Regime VII — Stochastic
With a stochastic constraint ε = 0.05·√N + 0.30·N·ξ_n (where ξ_n ~ N(0,1)), the system produces an ensemble of trajectories rather than a single path. The shaded band would show the 10th to 90th percentile range across 50 realizations; the solid line would show the ensemble mean. The spread of individual trajectories grows with n, reflecting the amplification of noise by the doubling map.
5.8 Regime Summary
| Regime | Constraint ε(N) | Parameters | Asymptotic behavior |
|---|---|---|---|
| I | 0 | — | N_n = 2ⁿ |
| II | κNᵅ | κ=0.05, α=0.5 | N_n ~ C·2ⁿ·(1 − O(n^{α−1})) |
| III | κN | κ=0.30 | N_n = (1.70)ⁿ |
| IV | κN/log N | κ=2.0 | N_n ~ C·nᵖ |
| V | threshold, ε₂=1.5N | N*=10⁶ | N_n → N* |
| VI | delay τ=2, ε=0.95N_{n−2} | — | growing oscillations |
| VII | stochastic, multiplicative | σ=0.30 | ensemble mean ~ Regime II, variance ~ 4ⁿ |
6. Relation to Existing Work
The constraint taxonomy introduced in this paper is not proposed as a replacement for existing frameworks in discrete quantum gravity and cosmology. It is proposed as a complementary classification language — a common vocabulary for comparing growth models developed in different traditions. This section situates the taxonomy relative to four existing frameworks.
6.1 Causal Set Theory
Causal set theory (Bombelli et al. 1987; Sorkin 1991; Henson 2006) models spacetime as a locally finite partially ordered set — a discrete collection of events connected by causal relations. The fundamental dynamical postulate is that the number of causal set elements grows through a sequential growth process in which new elements are added one at a time according to a probability measure that respects causality.
The sequential growth process in causal set theory is a stochastic recurrence — each step adds elements with a probability that depends on the current causal set. This places causal set dynamics squarely within Class VI of the present taxonomy, with the stochastic constraint ε determined by the transition amplitudes of the growth process.
The present taxonomy complements causal set theory in two ways. First, it provides a deterministic skeleton against which the stochastic dynamics can be compared — the mean trajectory of the causal set growth process is a constrained deterministic recurrence, and identifying its constraint class characterizes the large-scale behavior. Second, the taxonomy’s expansion regimes provide a classification of the possible large-scale geometries that different causal set growth measures can produce.
6.2 Causal Dynamical Triangulations
Causal Dynamical Triangulations (CDT; Ambjørn et al. 2000, 2012) constructs a path integral over discrete spacetime geometries built from simplicial complexes. The key dynamical input is the relative weight of different simplicial configurations in the path integral, which determines which macroscopic geometries emerge at large scales.
CDT is not a recurrence relation in the sense of the present taxonomy, but the volume of the emergent spacetime — the number of simplices at a given proper time — does satisfy an effective growth equation that has been measured numerically. In the four-dimensional CDT simulations of Ambjørn et al., the volume profile as a function of proper time follows approximately a semiclassical de Sitter solution, which corresponds to near-exponential growth in the present taxonomy’s terms — specifically, the reduced exponential regime with a base determined by the cosmological constant.
The present taxonomy complements CDT by providing an analytical classification framework for the volume growth profiles that CDT produces numerically. If different choices of CDT coupling constants produce different volume growth regimes — exponential, near-exponential, polynomial, or saturating — the taxonomy provides the language to characterize and compare them.
6.3 Loop Quantum Cosmology
Loop Quantum Cosmology (LQC; Bojowald 2001; Ashtekar and Singh 2011) applies the quantization methods of Loop Quantum Gravity to cosmological minisuperspace models, replacing the classical Friedmann equation with a quantum difference equation. The fundamental evolution equation in LQC is a recurrence relation on the volume eigenvalues of the quantum geometry, with the constraint arising from the quantum geometry of the loop quantization.
This is the existing framework most directly analogous to the present taxonomy. LQC’s fundamental equation is a constrained discrete recurrence of exactly the form N_{n+1} = f(N_n, N_{n−1}, …) studied here. The constraint in LQC arises from the area gap — the minimum nonzero eigenvalue of the area operator in Loop Quantum Gravity.
The present taxonomy complements LQC by providing a classification of the large-scale behavior of LQC recurrences across different quantization ambiguities and matter contents. Different choices of quantization scheme in LQC produce different effective recurrences; the taxonomy’s seven regimes provide a systematic way to characterize which large-scale behaviors are consistent with observation and which produce saturation, oscillation, or polynomial growth.
6.4 Discrete Dynamical Systems
The mathematical theory of discrete dynamical systems — including the study of iterated maps, recurrence relations, and difference equations — provides the general analytical foundation for the present taxonomy. The classification of fixed points, periodic orbits, and chaotic behavior in one-dimensional maps is well established (Devaney 1989; Strogatz 1994). The specific recurrence N_{n+1} = 2N_n − ε(N_n) studied here falls within this general framework as a one-dimensional map with a doubling nonlinearity.
The present work extends the general theory in one specific direction: it focuses on the large-n asymptotic regime rather than the local behavior near fixed points, and it classifies constraint functions by the global growth regime they produce rather than by the local stability properties of their fixed points. This is a different and complementary perspective. The local theory asks “what happens near N*?” The present taxonomy asks “what happens as n → ∞?” Both questions are necessary for a complete understanding of constrained discrete growth.
The taxonomy also extends the existing literature by identifying the near-exponential regime — Regime II in the expansion taxonomy — as a distinct class that lies between pure exponential growth and the more heavily constrained regimes. This regime has not received systematic attention in the discrete dynamical systems literature, presumably because it is asymptotically indistinguishable from exponential growth and therefore of limited interest from a pure mathematics perspective. From an applied perspective, however, the sub-leading corrections that characterize the near-exponential regime are precisely the kind of quantity that would be observable in any system where the leading-order behavior is otherwise a clean exponential.
References for Section 6
Ambjørn, J., Jurkiewicz, J., & Loll, R. (2000). A non-perturbative Lorentzian path integral for gravity. Physical Review Letters 85, 924–927.
Ambjørn, J., Jurkiewicz, J., & Loll, R. (2012). Causal dynamical triangulations and the quest for quantum gravity. In Foundations of Space and Time, Cambridge University Press.
Ashtekar, A., & Singh, P. (2011). Loop quantum cosmology: A status report. Classical and Quantum Gravity 28, 213001.
Bojowald, M. (2001). Absence of a singularity in loop quantum cosmology. Physical Review Letters 86, 5227–5230.
Bombelli, L., Lee, J., Meyer, D., & Sorkin, R. (1987). Space-time as a causal set. Physical Review Letters 59, 521–524.
Devaney, R. L. (1989). An Introduction to Chaotic Dynamical Systems. Addison-Wesley.
Henson, J. (2006). The causal set approach to quantum gravity. In Approaches to Quantum Gravity, Cambridge University Press.
Sorkin, R. (1991). Spacetime and causal sets. In Relativity and Gravitation: Classical and Quantum, World Scientific.
Strogatz, S. H. (1994). Nonlinear Dynamics and Chaos. Addison-Wesley.
7. Discussion
The taxonomy developed in this paper raises four questions that deserve explicit attention.
7.1 Which constraint classes admit continuum limits?
A fundamental question for any discrete framework is whether its large-scale behavior converges to a continuum theory as the step size goes to zero. In the present taxonomy, the answer depends on the constraint class.
Class I (no constraint) trivially admits a continuum limit: the recurrence N_{n+1} = 2N_n converges, under appropriate rescaling, to the ordinary differential equation dN/dt = N log 2, whose solution is exponential growth.
Classes II and III — power-law and logarithmic constraints — also admit continuum limits, since their constraint functions are smooth and the recurrence can be written as a forward difference equation that converges to a first-order ODE as the step size shrinks.
Class IV (threshold constraints) presents a more delicate case. If the threshold N* is fixed independently of the step size, it survives the continuum limit and produces a piecewise ODE with a genuine discontinuity at N*. If N* scales with the step size, the threshold may wash out in the limit, and the continuum theory may be smoother than the discrete one.
Class V (delayed constraints) does not admit a simple continuum limit in general. Delay-difference equations with fixed delay τ converge to delay-differential equations — a qualitatively different class of ODE that retains memory of a finite past interval even in the continuum. This memory is a genuine feature of the discrete structure that survives the limit.
Class VI (stochastic constraints) converges, under appropriate conditions, to stochastic differential equations of Itô or Stratonovich type. The specific form depends on the noise model and the amplification properties of the doubling map.
7.2 Which constraint classes produce scale invariance?
Scale invariance — the property that the system looks the same at different scales — is a desirable feature for many models of natural growth, since many observed large-scale structures display approximate scale invariance.
In the present taxonomy, scale invariance is most naturally associated with the near-exponential regime (Regime II). A near-exponential trajectory N_n ~ C·2ⁿ has the property that N_{n+k}/N_n → 2ᵏ for large n, independently of n — which is a discrete analog of scale invariance. The constraint introduces a scale (through the parameter κ) but does not break the large-scale doubling symmetry.
Class III (logarithmic constraints) also produces approximate scale invariance at large scales, since the logarithmic drag becomes negligible relative to the exponential growth. Classes IV and V break scale invariance explicitly — the threshold N* and the delay τ introduce preferred scales into the dynamics. Class VI introduces a statistical scale through the noise amplitude σ.
7.3 Which constraint classes remain computationally stable?
Numerical stability is a practical concern for any recurrence relation intended for simulation. In the present taxonomy, stability is determined by the amplification properties of the doubling map.
Classes I through IV are computationally stable for admissible constraint functions: the recurrence N_{n+1} = 2N_n − ε(N_n) with ε(N) < 2N is well-conditioned, and standard floating-point arithmetic handles the trajectories without difficulty for n up to several hundred steps.
Class V (delayed constraints) requires more care. The recurrence depends on τ + 1 previous values, and the memory requirement grows with τ. For large delays and strong constraints, the characteristic polynomial of the delay-difference equation may have roots near the unit circle, producing near-neutral stability and slow convergence to the asymptotic regime. Numerical integration of Class V systems benefits from higher-precision arithmetic at large n.
Class VI (stochastic constraints) is stable in the ensemble sense — the mean trajectory is well-behaved — but individual realizations diverge from the mean exponentially. The variance of N_n grows as σ² · (4ⁿ − 1)/3 for the pure doubling map, which means that for σ > 0 and large n, individual trajectories are numerically unreliable as representatives of the ensemble. Monte Carlo averaging over many realizations is essential.
7.4 Which constraint classes are of the greatest general interest?
Of the six classes, Class II and Class IV stand out for different reasons.
Class II is of particular interest because the near-exponential regime it produces is the natural mathematical setting for any system whose large-scale behavior is close to exponential but carries a small, permanent, and in-principle-measurable correction. Determining the correct exponent α for a given constraint mechanism — whether it is truly constant (α = 0) or grows slowly with system size — is generically the key open calculation in applying this class to any specific system.
Class IV is of particular interest because it provides the mathematical language for phase transitions in a growth process — any system that behaves one way below some characteristic scale and a qualitatively different way above it is a Class IV candidate, and identifying the location and sharpness of that transition is the natural first question to ask of such a system.
Class VI is of practical importance whenever a system’s deterministic predictions need to be checked for robustness: any conclusion drawn from a Class I–V model should, in principle, be re-examined under Class VI perturbation before being treated as a stable feature of the model rather than an artifact of assuming perfect determinism.
8. Conclusion
This paper has introduced a mathematical taxonomy for constrained discrete growth processes based on the recurrence relation N_{n+1} = 2N_n − ε(N_n). Six constraint classes — no constraint, power-law, logarithmic, threshold, delayed, and stochastic — produce seven qualitatively distinct expansion regimes: exponential, near-exponential, reduced exponential, polynomial, saturating, oscillatory, and stochastic. The taxonomy is summarized in Table 1 and illustrated numerically in Section 5.
The classification is deliberately mathematical rather than physical. It does not propose a specific physical theory, derive equations of motion for a specific system, or identify particular values of the constraint parameters with particular physical quantities. What it does is provide a common language — a set of precisely defined categories with known asymptotic behaviors — that can be applied to growth models across mathematics, physics, biology, and computer science.
As shown in Section 6, Causal Set Theory, Causal Dynamical Triangulations, and Loop Quantum Cosmology all involve constrained discrete growth processes that can be situated within the present taxonomy. The most direct parallel is with Loop Quantum Cosmology, whose fundamental recurrence relation involves a constraint arising from the area gap — a fixed geometric minimum modifying an otherwise smooth evolution equation, structurally the kind of object Class II’s boundary case or Class IV describes. Whether the LQC area gap in fact belongs to one of these classes, in the precise technical sense defined here, is a concrete question this taxonomy makes tractable, though answering it would require working through LQC’s specific difference equation directly.
One open problem follows directly from the taxonomy and is worth stating explicitly: the functional forms of ε(N) that are capable of producing genuine polynomial growth (Regime IV) need systematic characterization. Pure logarithmic constraints remain near-exponential for every fixed κ > 0; stronger forms (for example ε(N) = κN/log N) must be identified and their asymptotic exponents determined analytically rather than merely illustrated numerically.
The periodic table organized the elements before quantum mechanics explained them. Biological taxonomy organized life before evolution explained it. This taxonomy organizes constrained discrete growth processes independently of whatever physical theory, if any, ultimately explains a given instance of it — which is precisely what makes it usable as a common language across different physical frameworks rather than a claim tied to any one of them.
Updated: 6 August 2026