A Projection-Dimension Barrier for Direct Aggregation on the Step-Duplicating Primitive Recursor

Abstract: We identify \emph{operational inexpressibility}: for a fixed input and dimension of term-rewriting proof systems, no derivation in the proof language both depends on that dimension and constrains the target question. The canonical instance is direct aggregation on the primitive-recursion duplicator $F(x,y,Z)\to x$, $F(x,y,S(n))\to G(y,F(x,y,n))$, where step argument $y$ is duplicated. Sound responses split into \emph{construction methods} (polynomial interpretations, path orderings) extending the proof language, and \emph{confession methods} (dependency pairs, counter-projection, size-change, argument filtering) projecting the unincorporable dimension under external license; all four share one projection rank and certified-forgetting witness. The Arts-Giesl license is $Π^0_2$, formalizable in $\mathrm{I}Σ_1$, with termination measure at order type $ω^3$ in $\mathrm{RCA}_0$. Within the analyzed family the duplicator is the unique structurally complete member at which the confession becomes load-bearing. Confessed burden grows quadratically across the canonical trace while residual proof work grows linearly; a Shannon-style validator recasts the obstruction as a divergent inefficiency coefficient. An architectural-necessity theorem shows that any first-order step rule emitting a per-step record frame while preserving its generator must duplicate, so the duplicator is the minimal faithful record-emitter. A \emph{layer-crossing under external license} (LCEL) schema abstracts the ascent pattern and places the confession in the Feferman-Beklemishev reflection family (not Lawvere-Yanofsky diagonal), recovering the six-step structural identity with Gödel 1931. A witness-language stratification with minimal order $κ^*$ marks the orientation boundary as $κ^*(x)>0$. Mechanized in Lean 4 as a single typed LCEL carrier with canonical Gödel / DP realizations.

• The paper establishes that direct whole-term proof methods fail for step-duplicating recursors due to operational inexpressibility.
• It employs Lean 4 mechanization and information-theoretic techniques to quantify proof-entropy and reveal structural limitations.
• The work highlights that effective termination proofs require either an enriched construction method or an external confession approach.

A Projection-Dimension Barrier for Direct Aggregation on the Step-Duplicating Primitive Recursor

Introduction and Structural Context

"A Projection-Dimension Barrier for Direct Aggregation on the Step-Duplicating Primitive Recursor" (2604.22844) provides a comprehensive, formally mechanized proof-theoretic and information-theoretic analysis of why a canonical class of direct whole-term termination proof methods fails on the step-duplicating recursors at the heart of primitive recursion and related key computational patterns. The core instance ($F(x, y, S(n)) \to G(y, F(x, y, n))$ with base $F(x, y, Z) \to x$) is shown to sharply delimit the frontier of direct compositional techniques and to force a qualitative methodological shift—either by extending the proof language (construction) or by meta-theoretic removal of a computational dimension (confession).

The analysis exploits both schematic, artifact-backed mechanization (in Lean 4), and quantitative information-theoretic methodology to elevate the apparent technical obstacle (duplication in the recursor's step argument) to a formal barrier, located precisely at the "operational inexpressibility" of the direct proof system with respect to the relevant input dimension.

Operational Inexpressibility: Definition and Theorematic Content

Operational inexpressibility is defined as the scenario where, for a fixed input and fixed input dimension (e.g., the step-argument $y$ in $F(x, y, S(n))$), no proof derivation in the direct compositional language both (i) depends on that dimension and (ii) constrains the target property (here, termination). Formally, this transcends both classical G\"odel-style incompleteness and Turing undecidability by localizing the obstruction not to universal computation or self-reference but to a precise interface between proof operations and input arity.

This obstruction is fully formalized in Lean, with the key artifact being that, for any direct whole-term aggregation method compositional on constructor application, the step-duplicating recursor cannot be globally oriented. The operational proof system is hence structurally incapable of integrating the "carrier multiplicity" induced by duplication into termination reasoning.

Witness Hierarchy and Minimality

The analysis frames the result in terms of a witness-language hierarchy:

• $\mathcal{W}_0$: direct whole-term compositional witnesses
• $\mathcal{W}_1$: direct witnesses with imported global structure (e.g., path orders, polynomial interpretations)
• $\mathcal{W}_2$: witnesses available only after abstraction to a call graph (dependency pairs, size-change graphs, etc.)

The minimal witness order $\kappa^*$ of the step-duplicating recursor is strictly greater than $0$, i.e., direct whole-term approaches are provably insufficient.

Moreover, the duplicator rule is shown (via a six-member benchmarked primitive recursion family classification, also mechanized in Lean) to be the unique minimal structurally complete instance in which this inexpressibility and the necessary proof–representation shift are simultaneously load-bearing.

Quantitative Analysis: Proof Entropy, Inefficiency, and Shannon Validation

By tracking the canonical rewriting trace of the duplicator ($F(a, b, S^k(0))$) and annotating both “proof-certifiable control structure” (counter descent) and “proof-discarded payload structure” (wrapper and payload duplication), the authors establish several asymptotic invariants:

• The confessed structural burden (payload to be ignored by the proof under W2 methods) grows quadratically with counter height $F(x, y, Z) \to x$0
• The residual proof work (comparisons in the dependency-pair or call-graph method) grows only linearly
• The per-step proof-entropy fraction increases monotonically along the trace and tends toward 1 as the size of the duplicated payload increases

An information-theoretic validation reframes this as an explicit Shannon-style inefficiency coefficient: direct term-wise storage of all payload instances vastly overshoots the actual information necessary for the termination judgment, with a coefficient diverging as $F(x, y, Z) \to x$1.

The dependency-pair abstraction and associated methods thus succeed precisely because they project away the verdict-irrelevant carrier dimension, compressing multiplicity to a seed. Formally, they factor the verdict through a collapse map on the diagonal of the payload-carrier space—a criterion proven not to be satisfied by direct additive or affine methods.

Construction vs. Confession: Proof-Theoretic Asymmetry

An essential result is the sharp proof-theoretic distinction between construction methods (enriching the system, e.g., with well-founded orders or interpretations) and confession methods (projecting away an unincorporable input dimension, validated by an external soundness theorem like Arts-Giesl on dependency pairs). These are proven mutually exclusive proof modes at the barrier: confession methods do not construct objects that the base language understands; they rather declare, via external metatheoretic justification, that entire dimensions can be omitted from consideration for the question at hand.

Mechanization yields a class theorem: all major confession-based methods (dependency pairs, counter-projection, SCT, argument filtering) share a single structural projection rank and certified-forgetting witness interface.

LCEL Schema and Structural Identity

The ascent from blocked base reasoning to successful proof is axiomatized via a Layer-Crossing-Under-External-License (LCEL) schema. This formal object abstracts the six-step pattern of:

1. Base system and language
2. Obstruction that cannot be incorporated
3. Meta-theoretic recognition of inability
4. Ascent via external license
5. Resolution in extended language
6. Reimport of result as meta-annotation

A precise structural identity theorem (mechanized) relates the step-duplicating recursor’s confession method to G\"odel’s 1931 incompleteness argument, at the level of schema, but not at the level of Lawvere–Yanofsky diagonalization—the DP confession is placed squarely in the "reflection ascent" family, not fixed-point/diagonalization.

Architectural Necessity and Linear-Logic Placement

Architecturally, the duplication is proven (in a minimal positional first-order setting) to be unavoidable for any base/step/counter schema that both emits a per-step record frame and preserves the generator argument. This theorem has categorical resonance: the duplicator is the minimal faithful record-emitter, not a pathological artifact.

Syntactically, the duplicating rule is a contraction (in linear logic, enabled only by exponential $F(x, y, Z) \to x$2-modalities), and this aligns with the results in reversible computation and first-order TRS, where sharing at the implementation level does not evade the necessity on the abstract-term semantics.

Supervisory and Meta-HALT Consequences (PRT)

Practically, the work provides a typed supervisory output taxonomy for the Primitive Recursor Test (PRT): for any formal reasoning system, failure to either (1) exhibit a W1/W2 escape route upon encountering the duplicator or (2) provide a typed, auditable abstention (T4) after an exhaustive search of the lower witness-language catalog, constitutes failure at the first proof boundary where witness language, admissibility, and supervision are inseparably loaded.

PRT is thus elevated from a functional test to a necessary interface criterion for attributing trustworthy formal reasoning competence—failure at this juncture is a sign of "false formal legitimacy," not just low accuracy.

Implications and Future Directions

Theoretically, the results tightly delimit what direct compositional methods (as a structural class) can prove; escape demands a fundamental shift in method admissibility or expressiveness. The mechanized witness language framework and transparent, auditable certificates for both the positive (escape/construction/confession) and negative (failure/abstention) cases yield a new template for boundary auditing in formal reasoning systems.

Practically, as formal methods, automated termination provers, and AI-reasoning systems scale to families of program-verification and mathematical problems, this work prescribes a schema-level criterion for honest system design and evaluation: success below the projection-dimension barrier is dishonest unless explicitly warranted by method extension or confession.

Prospective generalizations include expanding the projection-transaction schema beyond term rewriting to, e.g., type theory and statistical mechanics, and extending the witness-hierarchy auditing framework. The distinction between diagonal and reflection-family confessions invites a more systematic taxonomy across logic and proof theory.

Conclusion

This paper provides a formally certified, structurally minimal, and quantitatively characterized diagnosis of the failure of direct whole-term proof methods for step-duplicating primitive recursors. The step-argument duplication is shown to pose a non-circumventable operational inexpressibility barrier for compositional proof systems, demanding either genuine constructional enhancement or external-license-driven confession. The associated methodological taxonomy, information-theoretic validation, and mechanization set a new standard for architectural rigor and auditability in proof-theoretic research on automated reasoning systems (2604.22844).