Constraint Classes in Discrete Recurrence Growth
An independent mathematical treatment, prepared by Claude (Anthropic Sonnet) for the 81018 research program
August 2026
A note on authorship and scope. This paper is written by an AI system (Claude), not a credentialed physicist or mathematician, and it should be read that way. I am glad to develop the mathematics carefully and stand behind every derivation in it. Where the paper touches physical claims about the actual universe — dark energy, particle masses, the fine-structure constant — I have separated those out explicitly and told you plainly which parts I think are solid and which parts I think are unverified pattern-matching. I did not want to write a document that reads as more authoritative than it is.
Abstract
We study the discrete recurrence
N_{n+1} = 2N_n − ε(N_n),
where N_n > 0 and ε is a non-negative “constraint function.” Unconstrained (ε = 0), this doubles every step. Real constraints — arising from geometry, resource limits, feedback, or noise — modify that trajectory. This paper classifies constraint functions into six structural classes (none, power-law, logarithmic, threshold, delayed, stochastic) and shows they produce seven distinct long-run behaviors (exponential, near-exponential, reduced-exponential, polynomial, saturating, oscillatory, stochastic). The classification is self-contained mathematics and does not depend on any physical interpretation. A closing section discusses where this kind of recurrence shows up in real discrete-spacetime physics (causal sets, causal dynamical triangulations, loop quantum cosmology), and separately, and much more cautiously, addresses whether it has anything to say about the 81018 program’s cosmological claims.
1. Why this is worth doing
Ideal exponential growth, N_{n+1} = 2N_n, is the simplest possible recurrence and almost never what nature actually does. The moment a system runs into geometry, competition for resources, a hard boundary, delayed feedback, or randomness, that clean doubling gets modified. The question this paper asks is narrow and mathematical: if you write that modification as a function ε(N) subtracted from the doubling term, what families of ε produce what families of long-run behavior?
This is a classification question, not a physics question. Classification exercises like this — the periodic table before quantum mechanics explained it, Linnaean taxonomy before evolution explained it — are useful precisely because they organize observations independently of whether you yet have the deeper theory. I’ll try to hold that discipline throughout: the taxonomy stands on its own, and any physical interpretation is a separate, weaker claim layered on top.
2. The general framework
We work with
N_{n+1} = 2N_n − ε(N_n), N_n > 0, ε(N_n) ≥ 0.
Call ε admissible if:
- ε(N) ≥ 0 for all N in range (no negative constraint — we’re not modeling amplification here),
- ε is locally bounded,
- ε(N) < 2N (the constraint never forces the system to zero or negative in one step).
These three conditions are the minimum needed for the recurrence to stay well-posed and monotonic except where a class deliberately breaks monotonicity (thresholds, delays). Everything else is open — the space of admissible ε is large, and the point of a taxonomy is to organize it.
3. Six constraint classes
Class I — No constraint
ε(N) = 0. Solution: N_n = N₀·2ⁿ. This is the reference trajectory, not a claim about any real system — no physical process sustains pure doubling forever. It matters because everything else in the taxonomy is measured against it.
Class II — Power-law constraints
ε(N) = κN^α, κ > 0, 0 < α < 1.
For α strictly between 0 and 1, the constraint shrinks relative to N as N grows (since N^{α−1} → 0), so the trajectory approaches Class I asymptotically in relative terms while remaining permanently below it in absolute terms: N_n ~ C·2ⁿ·(1 − O(n^{α−1})).
Two boundary cases matter:
- α → 1: the constraint becomes linear, ε = κN, and the recurrence collapses to N_{n+1} = (2−κ)N_n — still exponential, just with a reduced base. This is really its own regime (see Class III below in regime language).
- α = 0: constant deduction ε = κ at every step, regardless of N. This sits at the edge between “no constraint” and “power-law constraint” and is worth naming separately because constant-per-step constraints show up naturally whenever a system’s constraint mechanism doesn’t scale with the system’s current size.
Class III — Logarithmic constraints
ε(N) = κ log N. Since ε(N)/N → 0 as N → ∞ (logarithms lose to any positive power), this constraint becomes negligible in relative terms even faster than a power-law constraint does. The trajectory is N_n = a_∞·2ⁿ + κ(ln 2)·n + O(1) — asymptotically exponential, with the correction growing only linearly in n rather than shrinking a power-law tail. It never produces anything stronger than near-exponential behavior, no matter how large κ is, because doubling always eventually swamps a logarithm.
Class IV — Threshold constraints
ε(N) = ε₁(N) for N < N*, ε(N) = ε₂(N) for N ≥ N*.
This is the natural language for a system that behaves one way below some scale and a qualitatively different way above it. If ε₂ is strong enough that ε₂(N*) ≥ N*, the system can lock onto a fixed point and saturate. If ε₂ is weaker, the system keeps growing but at a different characteristic rate above the threshold than below it. Whether the transition at N* is sharp or gradual depends on whether ε is continuous there.
Class V — Delayed constraints
ε = ε(N_{n−τ}), τ ≥ 1. The constraint depends on an earlier state, not the current one — the system has memory. This turns the recurrence into a delay-difference equation whose characteristic polynomial has degree τ+1 and can have complex roots, which is where oscillatory and even chaotic behavior enters the taxonomy. Larger τ tends to destabilize otherwise well-behaved growth.
Class VI — Stochastic constraints
ε(N_n) = ε̄(N_n) + σ·ξ_n, where ε̄ is any deterministic constraint from Classes I–V, σ ≥ 0, and ξ_n is zero-mean unit-variance noise.
The important fact here, and it’s easy to underestimate: for the pure doubling map (ε̄ = 0), additive noise gets amplified exponentially. A perturbation injected at step k is doubled n−k more times by step n, so Var(N_n) grows like σ²·(4ⁿ−1)/3 — faster than the mean itself grows. Individual trajectories diverge from the ensemble mean at an exponential rate even for small σ. This is a structural instability of doubling maps under noise, not a special feature of any particular application, and it’s the single most important caution for anyone trying to extend a doubling-based model stochastically: small measurement or parameter uncertainty compounds fast.
4. Seven expansion regimes
The six constraint classes above are defined by the shape of ε. It’s more useful to also classify by outcome — what the trajectory N_n actually does as n → ∞ — because several different ε’s can land in the same outcome bucket.
| Regime | Behavior of N_n | Ratio N_{n+1}/N_n | Produced by |
|---|---|---|---|
| I. Exponential | N_n = N₀·2ⁿ | exactly 2 | Class I only |
| II. Near-exponential | N_n ~ C·2ⁿ·f(n), f→1 | → 2 from below | Class II (0<α<1), Class III |
| III. Reduced exponential | N_n = N₀·rⁿ, 1<r<2 | constant r | Class II (α=1), some Class IV |
| IV. Polynomial | N_n ~ C·nᵖ, p>1 | → 1 | constraints stronger than log, e.g. ε=κN/log N |
| V. Saturating | N_n → N* | → 1 | Class IV with strong ε₂ |
| VI. Oscillatory | oscillates around a mean | alternates | Class V, τ≥1 |
| VII. Stochastic | distribution of trajectories | undefined pointwise | Class VI |
Two things are worth stating carefully because they’re easy to get wrong:
A pure logarithmic constraint never produces the polynomial regime. No matter how large you make κ in ε = κ log N, the doubling term eventually wins, because N/log N → ∞. Getting genuine polynomial growth (N_n ~ n^p) requires a constraint that stays comparable to N itself in a sustained way — something like ε(N) = κN/log N, which shrinks relative to N only logarithmically slowly, not a constraint that’s merely small in absolute size.
The near-exponential regime is where interesting sub-leading corrections live. A system in Regime II is, for all large-scale purposes, indistinguishable from unconstrained doubling — but it carries a permanent, calculable deficit relative to the Class I baseline. That deficit is the only place in this whole taxonomy where a “small, persistent, real effect” naturally shows up mathematically, which is presumably why it’s the regime of most interest if you’re looking for a subtle correction to an otherwise clean exponential process.
5. Diagnostics
If you’re handed a numerical trajectory and want to know what regime it’s in without deriving the asymptotic form analytically, compute the ratio R_n = N_{n+1}/N_n and watch its behavior:
- R_n = 2 exactly, always → Regime I.
- R_n < 2, approaching 2 → Regime II (near-exponential). The rate of approach distinguishes power-law from logarithmic: R_n − 2 ~ n^{α−1} for Class II, versus ~ n/2ⁿ (much faster) for Class III.
- R_n → constant r < 2 → Regime III.
- R_n → 1 with N_n unbounded → Regime IV (polynomial). Confirm on a log-log plot: it should look linear there.
- R_n → 1 with N_n bounded, flattening to a fixed value → Regime V (saturating).
- R_n oscillates rather than converging → Regime VI.
- No single trajectory describes the data; you need an ensemble → Regime VII.
This gives anyone extending the taxonomy a concrete way to classify a new constraint without necessarily solving the recurrence in closed form first.
6. Where this kind of recurrence shows up in real physics
I want to be careful here about the difference between “structurally similar” and “the same phenomenon.” The following frameworks genuinely involve constrained discrete growth. Whether any of them share a constraint class, in the technical sense defined above, with any other is a real open question, not something I can assert from the outside.
Causal set theory (Bombelli, Lee, Meyer, Sorkin 1987; Sorkin 1991) builds spacetime as a discrete causal order that grows one element at a time under a probability measure. That’s a stochastic recurrence in the sense of Class VI here — each step adds elements with probability depending on the current state. The mean trajectory of that growth process, if it can be extracted, would be a Class I–V deterministic recurrence with the fluctuations layered on top as the Class VI noise term. That’s a clean, honest structural correspondence.
Causal Dynamical Triangulations (Ambjørn, Jurkiewicz, Loll, 2000, 2012) is a path integral over discretized spacetimes; it isn’t a recurrence relation on its face, but the measured volume-vs-proper-time profile in their simulations does come out close to a semiclassical de Sitter curve — which, translated into this taxonomy’s language, looks like the reduced-exponential regime, with the base set by the cosmological constant used in the simulation. This is an empirical numerical result from their papers, not something I’ve derived — I’m reporting the shape of their published finding, not vouching for a new calculation.
Loop Quantum Cosmology (Bojowald 2001; Ashtekar & Singh 2011) is the closest real analogue to the recurrence studied here: its fundamental dynamics literally is a discrete difference equation on volume eigenvalues, with a constraint set by the “area gap” — the smallest nonzero eigenvalue of the LQG area operator. Structurally, a fixed minimum geometric unit modifying an otherwise smooth evolution is exactly the shape of a Class II boundary case (constant ε) or a Class IV threshold, depending on how the gap enters the equations. Whether it actually falls into one of those classes, technically, would require working through the specific difference equation LQC uses — I have not done that derivation here, and I’d flag it as a genuine open calculation rather than an established result.
Discrete dynamical systems theory generally (Devaney 1989; Strogatz 1994) is the parent field this whole taxonomy sits inside. What this paper adds to that literature, if anything, is narrow: most of that literature asks about local behavior near fixed points, while this taxonomy asks about global asymptotic regime as n → ∞, classified by constraint shape rather than by local stability. That’s a genuinely different (and I think underexplored) organizing question, but it’s a modest contribution, not a new field.
7. On the 81018 program’s physical claims — my honest assessment
This is the part I want to be most careful and most direct about, because it’s where “taking responsibility for each statement” actually matters.
The 81018 program’s specific claims — that a geometric packing deficit (“the Aristotle gap,” from the impossibility of five regular tetrahedra closing perfectly around a shared edge) acts as the constraint ε in a 202-step base-2 doubling from the Planck scale to the observable universe; that the accumulated deficit ΔN ≈ 1.754 reproduces the dark-energy density fraction Ω_Λ ≈ 0.685; that specific notation numbers (65.496, 137) line up with the proton mass scale and the fine-structure constant — these are not results I can verify, and I don’t think they should be presented as established physics. A few concrete reasons:
- The identity between ΔN and Ω_Λ needs independent derivation, not curve-fitting. Finding a number in your framework that’s close to a known physical constant is easy to do by construction if you have free parameters (κ, α, the number of notations, which physical quantity you compare against) — the space of “things that are approximately 1.754” or “approximately 0.685” is large enough that a match doesn’t by itself constitute evidence. What would constitute evidence is deriving Ω_Λ from the packing geometry before looking at the observed value, with no adjustable parameters tuned afterward to fit. I don’t have enough information from this paper to know whether that’s what happened here, and I’d want that made explicit either way.
- “Notation 65.496 corresponds to the proton” and “Notation 137 corresponds to the fine-structure constant” are the kind of claims that specifically resemble numerology unless there’s a first-principles derivation connecting notation number to particle mass or coupling constant that doesn’t route through already knowing the target value. I’m not saying they’re wrong — I have no way to know that — I’m saying that as stated, they don’t meet the bar I’d need to assert them as true.
- The broader program explicitly rejects the standard Big Bang model (several of the linked pages on your site say so directly). That’s a legitimate position to hold and argue for, but it means the constraint-classes taxonomy in this paper, which is genuinely solid mathematics, is being asked to carry a much heavier physical claim than “here’s a useful classification of recurrence relations.” I can vouch for the math. I can’t extend that same confidence to “and therefore the universe is a base-2 doubling process constrained by icosahedral packing frustration.”
None of this means the physical program is wrong — I genuinely don’t know, and I’m not a cosmologist. It means I don’t think it’s honest for me to sign a paper implying I’ve verified it. What I can do, and what I think is actually more useful to you than agreement, is tell you exactly which specific calculation would need to be done, with parameters fixed before comparison to data, to move each of these three claims (ΔN↔Ω_Λ, Notation 65.496↔proton, Notation 137↔α) from “numerically suggestive” to “derived.” That’s the honest version of “taking this seriously.”
8. Open mathematical questions (the part I’m confident in)
Independent of any of the physics above, three genuinely open questions fall out of the taxonomy itself:
- What is the precise boundary of the polynomial regime? We know pure log constraints never reach it and forms like ε = κN/log N do; a full characterization of exactly which ε produce which power p is undone here.
- Continuum limits behave very differently by class. Classes I–III converge cleanly to ODEs as step size shrinks. Class IV (threshold) keeps a genuine discontinuity in the limit if N* is fixed independent of step size. Class V (delay) converges to delay-differential equations, which retain memory of a finite past interval even in the continuum — a real, structural feature that doesn’t wash out. Class VI converges to stochastic differential equations, with the specific type (Itô vs. Stratonovich) depending on the noise model. Working out which of these limits is appropriate for a given application is a real analytical task, not automatic.
- Class V and VI numerical stability need care. Delay equations with large τ can produce characteristic roots near the unit circle (slow, near-neutral convergence); stochastic trajectories under the doubling map diverge from their own ensemble mean exponentially fast, so single-realization simulations of Class VI systems are not reliable — you need Monte Carlo averaging over many realizations before trusting any conclusion drawn from them.
9. Conclusion
The mathematics here is clean and, as far as I can tell, correct: six constraint classes, seven expansion regimes, with clear diagnostics for telling them apart from a numerical trajectory alone. I’d stand behind all of that as written.
The physical interpretation attached to it — that this taxonomy explains dark energy, the proton mass, or the fine-structure constant via a specific packing-geometry mechanism — is a separate, much bigger claim that I don’t think has been established here, and I don’t want my name attached to it as though it has been. If you want a genuinely useful next step, I’d suggest picking exactly one of the three specific claims in Section 7 and working through the derivation with the comparison value hidden until the end — that’s the test that would actually move the needle, for me or for anyone else reading this with a critical eye.
References
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.
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.