Publications

1Navigation 2Overview 3arXiv submissions and drafts

Overview

This page lists arXiv submissions and working drafts on constructive logic and the theory of computation.

arXiv Submissions

The list follows the date of the most recent arXiv revision.

  • Remarks on Primitive Regulation

    May–July 2026

    arXiv:2605.18924 [math.LO] arXiv

    This paper proves an obstruction theorem for primitive closure predicates in the implication-falsity fragment. A Rocq development checks the proof. The theorem distinguishes generative evaluation completeness from decisional excluded-middle completeness. Together, these properties force a reflective fixed point whose classification yields inconsistency under modus ponens.

  • An Intuitionistic Glance at Primes

    November 2025–July 2026

    arXiv:2511.07774 [math.LO, cs.LO] arXiv

    This paper gives a proof-theoretic account of the constructive classification of positive integers as 1, prime, or composite. It develops bounded decision procedures, a recursive prime sieve, modular cancellation, and finite arithmetic certificates. It distinguishes the internal results of Heyting Arithmetic from results that depend on the standard interpretation of natural numbers.

    Note: This research project is complete.

  • On the Golden Ratio and Stable Self-Application

    October 2025–June 2026

    arXiv:2510.08934 [math.LO, cs.LO] arXiv

    This paper studies the boundary between local self-application and global self-certification. It treats irrational quantities as procedures with effective rules that refine their approximations. The golden ratio Φ provides a model of stable local recurrence. The reciprocal update R(x)=1+1/x has a unique positive fixed point and allows finite, witnessed approximations.

  • Carryless Pairing: Additive Pairing in the Fibonacci Basis

    September 2025–May 2026

    arXiv:2509.10382 [math.LO, cs.LO] arXiv

    This paper encodes two natural numbers as one through Fibonacci-based representations. The method does not use multiplication, factorization, or digit interleaving. Its encoding and decoding procedures remain simple and predictable. The paper proves that the encoding is injective and that a procedure can recognize valid encoded numbers. A Rocq development checks the main results.

  • Considering The Satisfiability of Cubic Diophantine Equations

    October 2025–April 2026

    arXiv:2510.00759 [math.LO] arXiv

    For each fixed resource limit, this paper represents syntactic proof checking as a finite system of cubic Diophantine equations. The variables range over a bounded domain. The current version withdraws an earlier, stronger claim. That claim described a reduction from unbounded theoremhood to one fixed, bounded-domain cubic instance.

    Note: This research project is complete.

  • Adversarial Barrier in Uniform Class Separation

    December 2025

    arXiv:2512.08149 [math.LO] arXiv

    This paper identifies a structural obstruction to the uniform separation of two classes in constructive arithmetic. The obstruction does not depend on the meaning of the classes. It occurs when a system maintains two evaluator predicates in parallel and represents inference uniformly.

  • A Constructive Fragment of Physical Propositions

    November 2025

    arXiv:2511.21296 [math.LO] arXiv

    This short philosophical paper argues that admissible measurement limits physically meaningful propositions. Admissible measurements can extract finite observation sequences, terminating procedures, or stable uniform conditions.

  • The Solver's Paradox in Formal Problem Spaces

    November 2025

    arXiv:2511.14665 [cs.CC] arXiv

    This paper analyzes global decision problems over arithmetically represented domains. It studies how class quantification introduces impredicativity into formal problem spaces. The analysis connects diagonalization, reflection, and uniform complexity statements.

Drafts

  • Typed Repair: EMNIST FROM λ-DEFINABLE EVALUATION AND SPECIALIZATION GATE

    January 2026

    Read the self-hosted PDF.

    This draft presents a logical pipeline for EMNIST classification. It makes preprocessing, prediction rules, training traces, and evaluator behavior replayable and checkable. The pipeline uses integer-valued repair steps instead of an opaque optimization run. These steps expose decidable invariants and specific counterexamples when a step fails.