One Trillion Possibilities, One Certified Boundary: The UNNS Investigation

Details
Written by: admin
Category: UNNS Research
UNNS Substrate Research Program · Structural Complexity Reduction · 2026

Does Structure Know Its Boundaries?

The Predictive Nature of the UNNS Substrate — how generative admissibility became a measurable physical boundary, a prospective prediction, and an exactly certified mathematical result.
Generative admissibility Exclusion gap Prospective prediction Lean-certified Finite model · 40 spins
Author: UNNS Research Collective Manuscript: v011 · 61 pages Evidence: commits 4e759d38 · 7610fe5d
Scientific illustration of UNNS structural predictability, showing a vast configuration space passing through a generative admissibility filter, a frozen 4×4 lattice prediction, an energetic exclusion boundary, and formal mathematical certification.
Fig. 1. From generative admissibility to certified structural prediction. The UNNS Structural Complexity Reduction program connects recursive structural selection, physical sufficiency, prospective boundary prediction, and exact formal certification. The illustration summarizes the progression from a vast nominal configuration space to a quantitatively tested and machine-certified exclusion gap.

What if the structure of a system could tell us something about its physical possibilities before those possibilities had been exhaustively examined?

This question lies near the heart of the UNNS (Unbounded Nested Number Sequences) Substrate. In its developing architecture, structure is not merely a collection of objects obeying externally imposed rules. Structure emerges through generation, acquires admissibility conditions, occupies realizability regimes, and follows routes whose boundaries influence what can persist or change.

But can this architecture do more than describe?

Can it predict?

The Structural Complexity Reduction (SCR) program was designed to confront that question directly. Could a generative structural description restrict an enormous nominal configuration space? Would that restriction remain physically meaningful? And could the boundary of its validity be predicted before the corresponding larger physical system was evaluated?

The investigation has now reached a decisive milestone. In a finite model of forty binary degrees of freedom — more than a trillion nominal configurations — a quantitative prediction about the boundary of generative admissibility was registered in advance, then verified by exact arithmetic and machine-checked proof.

Along the way the experiment uncovered something equally consequential: an extraordinarily restrictive grammar can fail to preserve physically relevant states — and that failure has a definite, measurable boundary.

4×4 nominal space
1.10 trillion
2⁴⁰ possible binary configurations
Generative space
193,662
grammar-admitted configurations
Certified boundary
0.945917
normalized exclusion gap, in kBT units
Formal certification
1,148
Lean proof batches accepted

IThe difference between possible and generable

Mathematics is generous with possibility. Forty two-state spins can be arranged in 2⁴⁰ ways — over a trillion nominal configurations. But a generative grammar is not generous. It is a rule for how structure may be built, and it admits only what the rule can produce.

The UNNS proposition is that these two sets — the nominally possible and the generatively admissible — need not coincide, and that the gap between them is itself a structural fact. In the test system, a free-boundary square spin-ice lattice, the grammar is the local vertex condition known as the ice rule. Imposed on a 4×4 patch, it admits 193,662 configurations out of 2⁴⁰: a contraction of 99.99998 %.

Numerous glowing lattice configurations converge through a structural admissibility gateway, emerging as a much smaller, organized family of admissible states.
Fig. 2. Generative admissibility contracts the nominal space of possibilities. A frozen structural grammar restricts which configurations can be generated before physical energy is evaluated. In the 4×4 square-ice model, 193,662 configurations satisfy the grammar out of 2⁴⁰ nominal states, illustrating the enormous difference between mathematical possibility and generative admissibility.
Contraction by the structural grammarLogarithmic bars comparing nominal and grammar-admitted configuration counts for the 3×3 and 4×4 lattices.Contraction of the configuration space (logarithmic scale)100102104106108101010123×3 · 24 spins2²⁴ = 16,777,216 nominal2,604 admitted (÷ 6,443)4×4 · 40 spins2⁴⁰ ≈ 1.10 × 10¹² nominal193,662 admitted (÷ 5,677,477)
Bar lengths are logarithmic (powers of ten). The 3×3 figure comes from a constructive run; the 4×4 figure is an exact combinatorial count — a count, not a Lean result.

Contraction on this scale is remarkable. But contraction alone proves nothing about physics. Whether a small family of generated structures is the right family depends on what you are trying to represent — and that is where the investigation turned.

IIThe unexpected lesson: structural selection is not physical preservation

The natural hope was that the ice rule — a rule the low-energy physics of this material is famous for obeying — would keep exactly the states that matter. A preregistered test on the 3×3 patch (24 spins) put that hope to work. The grammar admitted only 2,604 of 16,777,216 configurations. The physical target was defined independently of the grammar: every state within one thermal energy unit (1 kBT) of the ground level.

That window contains 26 states. The grammar kept 10. It excluded 16.

Glowing lattice configurations pass through a structural selection gateway toward a low-energy landscape, illustrating the distinction between grammar-admitted states and an independently defined physical target.
Fig. 3. Structural selection and physical sufficiency are distinct properties. Strong generative contraction does not guarantee preservation of physically relevant configurations. In the 3×3 experiment, the grammar admitted only 2,604 of 16,777,216 nominal configurations, but excluded 16 of the 26 states within the registered thermal energy window. This result establishes why UNNS admissibility must be evaluated against an independently defined physical target.
Sixteen of twenty-six target states excludedDot grid: 10 states preserved and 16 excluded in the 3×3 one-k_BT window.The 3×3 energy window: 26 target states10 inside the grammar (preserved)16 excluded by the grammarWindow = 1 kBT above the ground level · 16 of 26 = 61.5 % of the physical target lies outside the grammar
Each circle is one state in the 3×3 one-kBT window: 10 inside the grammar, 16 outside.

Generative boundary

What the grammar can produce. It is defined by the rule alone, before any energy is computed.

Physical relevance

What the target contains. It is defined independently, by an energy window, and does not consult the grammar.

This is a negative result for the simple hypothesis — and the most instructive moment in the program. A grammar can be extremely restrictive and physically insufficient. The two properties are distinct, and a structural description must always be evaluated against a target it did not choose.

IIIDiscovery of the exclusion boundary

The failure was not random. The excluded states were not scattered through the window; the very first one entered at a precise energy. Call it the exclusion gap, Δ: the distance in energy from the ground state to the lowest-lying configuration the grammar cannot generate.

A grammar may then represent a low-energy window completely, provided the window is narrower than Δ — and become insufficient the moment the window widens past it. The boundary of physical sufficiency stops being a vague judgement and becomes a number.

The gap equivalence
ST(W) ⊆ SG  ⇔  W < Δ¬G
The grammar contains the entire physical target for an energy window W exactly when W is smaller than the exclusion gap. A generative description has a measurable domain of physical sufficiency. The exclusion gap identifies its exact boundary.
Two glowing spin-lattice configurations occupy different levels of an energy landscape, with a highlighted corner defect in the upper configuration and a positive energy gap separating it from the ground-state basin.
Fig. 4. The exclusion gap identifies the energetic boundary of physical sufficiency. The lowest grammar-excluded configuration lies a strictly positive energy above the global ground state. In the certified 4×4 model, this threshold is approximately 0.945917 kBT and is attained by a configuration containing a localized corner violation. Below the exclusion threshold, the grammar preserves the complete energy-window target; at and above it, preservation fails.
Gap equivalence on the window axisNumber line showing the certified boundary at 0.945917 k_BT separating sufficiency from insufficiency.The sufficiency boundary on the energy-window axisW < Δ : the grammar containsthe whole physical targetW ≥ Δ : a grammar-excludedstate has entered the window0.50.7511.251.5energy window W (units of kBT)Δ ≈ 0.945917 kBTthe registered 1 kBT window lies on the right-hand side
The certified 4×4 boundary on the energy-window axis. The 1 kBT window used in the experiment sits just beyond it.

An exact criterion, not a shortcut

The equivalence tells us what it means for a grammar to be sufficient. It does not by itself tell us the value of Δ for a system we have not yet solved. That is the next step: can Δ be foreseen?

IVThe decisive step: predicting before calculating

The gap had been measured for the 1×1, 2×2 and 3×3 patches. Those values — 0.871, 0.952 and 0.943 in units of kBT — formed the evidence base. From them a prediction for the 4×4 patch was frozen and registered before the 4×4 system was evaluated. It had two parts:

  • Direction: the 4×4 normalized gap δ₄ lies below 1 — so a 1 kBT window must already breach the grammar.
  • Range: δ₄ lies in the interval [0.7904, 1.0334], derived from the smaller patches.
A glowing structural prediction rule leads through four progressively larger square lattices, labeled 1×1, 2×2, 3×3, and 4×4, ending with a certified result.
Fig. 5. Predicting a structural boundary before evaluating the larger system. Previously examined lattice sizes provide the basis for a frozen quantitative prediction of the 4×4 normalized exclusion gap. The subsequent exact result, δ₄ ≈ 0.945917, satisfies both registered predictions: δ₄ is below 1 and lies within the specified interval. This prospective test demonstrates structural predictability in the investigated finite physical model.
Exclusion gap by lattice sizeLine chart of the normalized exclusion gap for 1×1 to 4×4 with the registered prediction interval and the δ=1 line.Normalized exclusion gap δ_L across lattice sizesregistered interval R₂ = [0.790, 1.033]0.80.91.0δ = 1 : window equals the gapδ₁ = 0.8714111×1 · basisδ₂ = 0.9523822×2 · basisδ₃ = 0.9428983×3 · basisδ₄ = 0.9459174×4 · predicted, then certified
The measured sequence δ₁…δ₃ forms the basis; δ₄ = 0.945917 falls inside the registered interval and below the δ = 1 line.

The result: δ₄ = 0.945917. It satisfied both parts.

Why does that change the evidentiary standing of the study? An explanation fitted after the fact can always be made to work. A prediction fixed in advance, from smaller systems, can fail — and the 4×4 result could have landed outside the interval or above 1. It did neither.

What this does — and does not — show

The registered interval is broad, and the directional prediction (δ₄ < 1) is a modest one. The planned feasibility campaign was not completed, and an earlier exploratory verdict was inconclusive. The claim is therefore calibrated: structural predictability is demonstrated for this finite model, with a wide interval — not a precision forecast, and not yet a general law.

Single-reversal energy tiersDot plot of 40 single-island reversal energies in three tiers, with the 1 k_BT window.Forty single-spin-island reversals from one certified ground witness0.81.01.21.41.61.82.08 at 0.9459 kBT — corner vertex (inside the window)8 at 1.048924 at 1.79–1.911 kBT window ◂ insideenergy above the ground level (kBT) · each dot = one island reversed
Where the boundary lives in the 4×4 lattice: the cheapest grammar-breaking reversals are the corner vertices, at the gap value itself. This is a calculation on one certified ground witness, not an optimisation over every possible defect.

VWhen computation became mathematical certainty

A number from an optimizer is only as trustworthy as the optimizer. So the program went one step further. Using exact arithmetic — the 780 couplings of the model encoded as dyadic rationals — a branch-and-bound search was reformulated so that every pruning step could be checked by a proof assistant. The proof was assembled in Lean 4 from 1,148 independently kernel-checked proof-tree batches: 344 for the ground state and 804 for the lowest grammar-excluded state.

An illuminated branching proof tree feeds exact arithmetic certificates into a Lean verification kernel, producing verified global energy minima for the complete and grammar-excluded configuration spaces.
Fig. 6. From numerical optimization to machine-checked mathematical certainty. SCR-IV uses exact arithmetic, formally justified branch-and-bound pruning, and Lean kernel verification to certify the global ground energy and the minimum energy among grammar-excluded states across the 2⁴⁰-state configuration space. The resulting proof establishes a positive finite exclusion gap and verifies both prospective prediction inequalities for the encoded Hamiltonian.
Lean proof batchesStacked bar of 344 ground-state and 804 excluded-state proof batches.1,148 independently kernel-checked proof-tree batches344 · ground state804 · lowest grammar-excluded statetwo exact minima → a strictly positive gap → both registered predictionsΔ = 12 095 447 728 404 694 790 / 2⁷⁰ A (exact, dyadic)
The certified gap is an exact dyadic rational, expressed in units of the coupling scale A.

The certificates establish both global minima, a strictly positive gap between them, and both registered inequalities — for the whole encoded 40-spin model, independently of the numerical search.

A correction that changed nothing — and showed the method working

The certification also revealed that the lowest grammar-excluded energy originally reported sat 928/2⁷⁰ A (about 7.9 × 10⁻¹⁹ A) above the true minimum. The proof replaced the minimizer, moved δ₄ by less than 10⁻¹⁵, and altered no verdict. The original record was left untouched; the correction was added beside it. Verification is not ornamental — it can find what a solver, working at its own resolution, could not distinguish.

Explore the Evidence: The Structural Predictability Dashboard

The investigation is accompanied by an interactive research dashboard bringing together generative contraction, physical preservation, exclusion-gap prediction, exact certification, Hamiltonian sensitivity, and the underlying archival evidence. Readers can explore how the physical energy window changes the sufficiency of a structural grammar and trace the results to the frozen manuscript and research records.

The six images explain the scientific ideas visually. The dashboard lets readers examine the findings and their provenance.

Open the dashboard in a separate window ↗ Read the manuscript (PDF)

Together with the manuscript PDF and fixed GitHub commits, the dashboard completes the presentation of the Structural Predictability investigation.

VIWhat this reveals about the UNNS Substrate

Strip away the lattice, and what remains is a statement about the substrate itself. UNNS describes structure as something that is generated, that must be admissible to be realized, that follows routes, and that has boundaries beyond which it does not persist. SCR asks whether one of those boundaries can be turned into a measurable, forecastable quantity. In this finite model, it can. Five insights follow.

Insight 1

Generation and physical realization are distinguishable

The possibilities produced by a grammar are not automatically identical to the configurations relevant to a physical system. 16 of 26 target states lay outside.

Demonstrated
Insight 2

Admissibility has quantitatively testable boundaries

A generative description may be sufficient over one domain and insufficient beyond a definite threshold: W < Δ.

Demonstrated
Insight 3

Boundaries can become predictive variables

Rather than predicting only physical states, we can investigate when a structural description will cease to encompass a designated target.

Demonstrated in the finite model
Insight 4

Structural properties can be related through margins

The exclusion gap is an energetic margin that may be studied alongside connectivity margins, rigidity regions and boundary-routing behaviour.

Opens research — not established
Insight 5

Prediction can be certified

Formal verification establishes boundary claims as exact finite-model statements, independent of the numerical optimizer.

Demonstrated
The central statement
A generative description has a measurable domain of physical sufficiency. The exclusion gap identifies its exact boundary.

The last two insights deserve care. That the gap is a margin that could sit beside the UNNS connectivity margins is a research direction, not a finding: no relation between these quantities has yet been shown. What has been shown is narrower and firmer — that this kind of boundary can be defined, forecast from smaller systems, and proved.

The significance is not confined to the particular lattice used as the test environment. It concerns the developing predictive character of structural admissibility itself.

VIIThe next frontier: predicting across domains

Honest accounting separates three levels of achievement:

Level 1

Structural predictability

A boundary forecast from smaller systems, in a finite model.

Demonstrated
Level 2

Predictive verification

The forecast confirmed independently of the search, by Lean, for the encoded model.

Achieved
Level 3

Cross-domain predictability

The same procedure transferring to systems with no spin-ice in them.

Not established — the next test

The spin-ice model supplied the experiment. The question it leaves is about the substrate: whether connectivity margins, rigidity, canonicality and structural routes can help predict sufficiency boundaries in other systems. A fair test would require five things: a grammar frozen blind to the target; an independently defined target; a measurable exclusion quantity; a prediction frozen from smaller instances, with an interval narrower than the spread of what it was built from; and verification independent of the search.

That is a proposal, not a result. It is also the shape of a falsifiable program.

Beyond the experiment: toward the predictive geometry of the UNNS Substrate

The Structural Predictability investigation has given the UNNS Substrate a new kind of measurable object: a boundary separating the domain in which a generative description remains physically sufficient from the domain in which it does not. This result joins a growing body of UNNS research into admissibility, realizability, connectivity, rigidity, canonical structures and structural motion.

Together, these investigations suggest an increasingly precise question about the substrate. Are its different structural boundaries — those governing generation, connectivity, persistence, rigidity and physical sufficiency — related through a deeper organization that can itself be predicted?

Structural Complexity Reduction has not answered that entire question. It has established a rigorous route by which one part of it can be investigated.

A generative rule was frozen. A physical target was defined separately. A boundary prediction was registered before evaluation. The resulting finite-model boundary was established through formal mathematical proof.

Structural admissibility has become an experimentally testable source of quantitative predictions. The next challenge is to discover how far that capability extends — and whether the substrate's different forms of structural organization reveal common principles governing the emergence, persistence and transformation of physical structure.

The investigation began by asking what structure excludes. It has brought us to a more fundamental question: what can structure predict?

Manuscript & evidence links

UNNS Substrate Research Program · Structural Complexity Reduction · UNNS Research Collective · Finite-model results; cross-domain transfer is not established. The six illustrations are conceptual, not plots of the numerical data; the diagrams in the text are drawn directly from the manuscript's frozen values. unns.tech

Main Menu