Work

Positive Seeds, Exactness, and Certificates for Operation Search

Year: 2026
Type: preprint
Abstract: This paper studies how starting constants, strict evaluation domains, and representation choices determine what an operation language can express. Its continuous example is F(a,b) = exp(a) − log(b), with b > 0 required at every operation node. A fixed four-node expression gives a constructive proof that every positive starting constant generates a dense set of real values. The paper derives convergence bounds and a criterion for uniform approximation on compact input sets. It also proves that the positive constants capable of generating 1 exactly are dense and satisfy a dichotomy: either every positive constant has this property, or only countably many do. Further consequences from o-minimality describe finite expression budgets and the geometry of exactness with several named constants. For finite operation languages, attainable output columns form the least relation closed under the declared operations. This yields complete impossibility certificates on a full finite input domain and precise conditions for transferring exclusions between representations. Exact experiments study Boolean obstructions and matrix-generator costs, including a parameterized optimality certificate checked through 49,152 symbolic Bellman inequalities. The accompanying reproducibility supplement contains 15 compiled mathematical Lean modules, pinned dependencies, an axiom audit, executable experiments, and scoped negative-memory records. The manuscript distinguishes Lean-checked results from ordinary mathematical proofs and finite computational evidence. In particular, the o-minimal extensions and specialized matrix-cost tables are not presented as fully formalized in Lean. Whether the starting constant 2 can generate 1 exactly, which branch of the exact-seed dichotomy holds, and whether the displayed arithmetic expressions have minimum size remain open.
Language: English
Subfield: Software
Open Access status: green
OpenAlex ID: W7235405846
Cites:
Cited by:
Related to: 10