A variation of Reynolds-Hurkens Paradox

Thierry Coquand

DOI: https://doi.org/10.48550/arXiv.2308.16726

2023-08-31

Abstract:We present a variation of Hurkens paradox, which can itself be seen as a variation of Reynolds result that there is no set theoretic model of polymorphism.

Logic in Computer Science,Logic

What problem does this paper attempt to address?

The problems that this paper attempts to solve are about variants of certain paradoxes in type theory and their encoding and computational behavior in higher - order logic and specific systems (such as λU−). Specifically, the paper explores how to express and analyze these paradoxes in these systems and reveals some limitations of existing type systems.

### Main problems:
1. **Expression and Encoding of Paradoxes**: The paper discusses a variant of Hurkens' paradox, which itself is a variant of Reynolds' "paradox". The author shows how to express these paradoxes in higher - order logic (HOL) and the system λU−.

2. **Requirement for Extension of Type Systems**: In order to analyze the computational behavior of these paradoxes, the author believes that existing type systems need to be extended, especially by introducing the concepts of definitions and head linear reductions. This helps to better understand the computational characteristics of these paradoxes.

3. **Computational Behavior of Paradoxes**: The paper describes in detail the computational behavior of these paradoxes in different systems, including the behavior of looping combinators and whether they can be reduced to themselves.

### Specific Contents:
- **Background of Paradoxes**: The paper first introduces a variant of Russell's paradox and explains its manifestation in higher - order logic. Then, the author shows how to express this paradox in the λHOL system and proves a theorem, stating that in λHOL, there cannot be a type A such that Pow(PowA) is a judgemental retract of A.

- **Improvement of Paradoxes**: The author further improves this paradox by introducing two mappings, intro and match, and assuming that they satisfy a certain convertibility condition. In this way, the author obtains a new paradox and proves that in λHOL, there cannot be a type A and two mappings intro: T A → A and match: A → T A such that match◦intro is convertible to T(intro◦match).

- **Encoding in λU−**: The paper shows how to encode these paradoxes in the system λU− and discusses the computational behavior of this encoding. The author points out that although these paradoxes do not directly reduce to themselves, they will exhibit cyclic behavior.

- **Definitions and Computational Behavior**: The author emphasizes the importance of definitions (let - expressions) and head linear reductions in understanding the computational behavior of these paradoxes. Especially in dependent type systems, definitions cannot be simply replaced by abstraction and application because they play an important role in type checking.

### Conclusion:
The main contribution of the paper is to show how to express and analyze certain types of paradoxes in higher - order logic and the λU− system and to point out the limitations of existing type systems. By introducing the concepts of definitions and head linear reductions, the author provides a new method to understand and handle the computational behavior of these paradoxes.