81018Paper II — Doubling Constraints

MS-2026-08-07Draft

Meta AI 81018 Research ProgramAugust 7, 2026·Paper II

Constraint Classes for the Doubling Map: A Formal Asymptotic Taxonomy

Manuscript independently derived by Meta AI, August 2026. Released for 81018.com as Paper II of the synthetic peer review constraint program.

Authors

Meta AI 81018 Research Program

Paper II — Doubling Constraints

N₀ = 100, T = 32 steps, ε ≥ 0

Citation

“Growth is not denied, it is taxed. The tax function ε determines the rate.”

Abstract

We study perturbations of the pure doubling map  under formal constraints. Writing perturbed dynamics as  with , we classify possible asymptotics via syntactic form of  and semantic behavior of ratio . The taxonomy distinguishes seven canonical regimes ranging from negligible sublinear drain to stochastic and delayed feedback, with sharp thresholds for , , and . A Master Lemma unifies the mapping from syntactic cost to semantic growth.

1. Framework

Let  denote a resource measure at discrete time . The idealized law is doubling:

Constrained versions take the form , where  is a cost, drain, or friction functional. Define the instantaneous multiplier

Classification reduces to the order of . If  then . If  then . If  then . This trichotomy is syntactic in  but semantic in .

2. Syntactic Classes — Seven Canonical Forms

I

Pure Doubling

exponential baseline — unperturbed doubling, the reference clock.View

II

Sublinear Correction drain

negligible tax: sqrt-drain is absorbed; asymptotically pure.View

III

Subdoubling Linear

constant factor <2 — proportional overhead, reduced base.View

IV

Logarithmic Brake

polynomial-like: 

, growth is 

.View

V

Threshold / Saturation

piecewise: 

, capacity-limited doubling.View

VI

Delayed Feedback

memory term, characteristic 

, 

 oscillating.View

VII

Stochastic Perturbation

mean-preserving noise, log-normal around 

.View

3. Semantic Regimes — Master Lemma

Lemma — Unified Growth TrichotomyMaster Lemma 3.1

Let , , .

(i)

Negligible: If 

 then 

 and 

.

(ii)

Proportional: If 

 then 

.

(iii)

Saturating: If 

 then 

 and growth is subexponential.

Proof sketch

, limits follow directly. Trichotomy exhaustive for  monotone. Delayed and stochastic replace  with history/noise; same criterion holds pathwise or in expectation. Delayed case reduces to root analysis of  with dominant root . Stochastic preserves  under mean-preserving multiplicative noise.

R → 2

Classes I, II, VII (mean)

asymptotically pure doubling, o(N) error

R → c ∈ (1,2)

Classes III, V(late), VI

exponential with reduced base

R → 1

Class IV, V(capped)

polynomial-like, log-braked

4. Mapping Table — From Syntax to Semantics

Class			Growth	Form
I Pure Doubling	0	2.00	2ⁿ	exponential baseline
II Sublinear Correction	0.05√N	2⁻	Θ(2ⁿ)	o(N) drain
III Subdoubling Linear	0.3N	1.70	1.7ⁿ	constant factor <2
IV Logarithmic Brake	N(1-2/log(2+N))	1.00	exp(Θ((log N)²))	log-brake
V Threshold/Saturation	0→0.85N	2→1.15	2ⁿ→1.15ⁿ	capacity-limited
VI Delayed Feedback	0.95 Nₙ₋₂	≈1.68 osc	Θ(ρⁿ),ρ≈1.68	memory term
VII Stochastic	random	2±σ	log-normal around 2ⁿ	stochastic

Table 1: Canonical mapping. All asymptotics for N₀=100 unless stated. ε(N) in monospace denotes syntactic cost. Scroll horizontally on mobile to view full table.

5. Canonical Examples — Trajectories

I · baselineII · o(N) drainIII · c=1.7IV · log-brakeV · capacityVI · memoryVII · noise

Selected

II · Sublinear Correction

N₀=100 · 32 steps

log₁₀ NₙlogNRₙ

n=0 → 32

N₃₂ ≈ 4.26e+11 · R̄ ≈ 2.00

ε(N)

0.05√N

R∞

2⁻

Growth

4.3e+11

Sublinear correction shows negligible drain.  is , so . Visually indistinguishable from pure doubling after , yet formally non-zero. Models GC, small leak, or sampling noise.

Interpretation

 ⟺ growth rate = ideal − relative cost. Taxonomy is syntax → semantics mapping for any system claiming doubling.

6. Relation to Frameworks

Amortized Analysis

Sublinear  = negligible overhead; proportional  = constant-factor slowdown. Classes I–II are  amortized vs doubling; Class III is .

Population Dynamics

Classes IV–V are logistic-like with carrying capacity in growth rate, not absolute size. Brake IV is , .

Distributed Systems

Threshold V models shard saturation; delayed VI models feedback lag in autoscaling.  oscillation is health-check delay signature.

Stochastic

Class VII preserves doubling in expectation but log-normal variance grows linearly: .

7. Open Problems

1

Sharp delayed threshold

For N_{n+1}=2N_n – α N_{n-k}, determine critical α_k at which ρ(α_k) crosses 1. Conjecture α_k = 2^k/(k+1)^{k+1} scaling. What is exact oscillation frequency at threshold?

2

Universality of logarithmic brake

Classify all ε(N)=N·f(N) with f(N)→0 slowly such that log N_n ~ (∫ f^{-1})? Show Class IV is universal attractor for f(N)=c/log N + o(1/log N) up to time change.

3

Stochastic concentration

For Class VII with mean-preserving ξ_n, prove uniform concentration N_n / E[N_n] → exp( Gaussian ) with optimal Berry-Esseen rate. When does median overtake mean? Relation to doubling vs typical?

4

Composition and tensoring

If ε = ε₁ ∘ ε₂ or ε(N₁,N₂) = ε₁(N₁)+ε₂(N₂) on product systems, does R∞ factor? Study monoid of constraint classes under sequential and parallel composition. Is taxonomy closed?

8. Conclusion

The constraint taxonomy is a compact lens. Any system that claims doubling must declare its . If , the claim stands. If , the claim is subdoubling. If , the claim is braked. Delay and noise do not change the criterion, only the path. This unifies amortized, population, distributed, and stochastic readings of “almost doubling” into a single syntactic-semantic table.

Seven syntaxes, three semantics, one equation: .

#doubling-map#constraint-classes#asymptotic-taxonomy#formal-methods#synthetic-peer-review#81018#Meta-AI

Manuscript independently derived by Meta AI, August 2026. Released for 81018.com

Parent: constraint-classesMeta paper

Typeset with MathJax 3 · Charter / Fraunces / Newsreader · N₀=100 reference implementation · Draft v0.2.1 — No feedback loop, eight independent papers intended.