/topology

Formal topology (and some probability) in Coq

Primary LanguageCoqMIT LicenseMIT

A programming language for topology and probability in Coq.

Building

This project is aimed at Coq version 8.6. To build, run make at the base level of the project directory.

Files

Spec/

The "specification" for the continuous programming language, stated in terms of categories.

  • Category.v : Definition of categories and their properties
    • Cartesian monoidal categories (finite products)
    • Strong monads (e.g., measures and probability distributions)
  • Discrete.v : Discrete spaces
    • Every Coq type has an according discrete space, and every Coq function is a continuous map between the according discrete spaces
  • Sum.v : Sum spaces
    • The empty space, and binary sums (disjoint) unions of spaces
  • Lift.v : Lifted spaces
    • Given any space, adjoints a "bottom" element, which can be thought of as indicating non-termination. The bottom element is a generic point. Lifted spaces are compact because the new open set for the whole space must be in every open cover. This allows interpretation of general recursion.
  • Sierpinski.v : The Sierpisnski space
    • The Sierpinski space is homeomorphic to the lifting of unit. Perhaps I will end up defining it this way, but here it is specified as something on its own.
  • Stream.v : Infinite streams
  • Real.v : Real numbers
    • lower real numbers, non-negative lower-real numbers, non-located real numbers (upper and lower Dedekind cuts which may have an entire interval as a gap rather than just a point), and bona fide real numbers
  • Prob.v : Measure and probability spaces
    • Definition of open sets and abstraction from maps to the Sierpinski space
    • Measure, subprobability, and probability monads
    • way-underspecified coinflip distribution and normal distribution
    • probabilistic infinite streams

FormTopC/

Computationally relevant definitions of formal topology and some constructions.

Numbers/

  • Qnn.v : Non-negative rational numbers

    • semiring (0, 1, addition, multiplication)
    • truncated subtraction
  • LPReal.v : Non-negative lower real numbers encoded as lower Dedekind cuts

    • semiring (0, 1, addition, multiplication)
    • indicators of logical propositions
    • supremum, min, max

Algebra/

  • FrameC.v : Computationally-relevant definitions of preorders, partial orders, semilattices, lattices, and frames
  • SetsC.v : Computationally-relevant definitions of subsets and notations
  • Sets.v : Computationally-irrelevant definitions of subsets and notations

Base directory

  • Samplers.v : Random samplers

    • Definition of random samplers, proofs and constructions
  • Valuation.v (old, and full of lies!)

    • definition of valuations and continuous valuations
    • definition of simple functions, integration, and assumption of many facts about integration
    • operations for construction valuations: unit, bind, join, map, product, restrict, inject
    • attempted definition of measurability
    • supremum and fixpoint valuations, continuity of valuation functionals
    • principles for constructing and reasoning about finite and countable measures
    • examples: probabilistic choice, Bernoulli, binomial, geometric
    • example of Geom/Geom/1 queue system
  • Sample.v : Definition of random samplers

    • Samplers of the form R -> R * A, where we sample random values of A from a random seed R
    • Probability distributions over streams
    • Partial computations, partial valuations, and partial samplers
  • PDF.v : (Very incomplete) characterization of PDFs of measures relative to more standard measures

Types/

Facts about types. In particular, facts about isomorphisms/equivalences of types, and characterization of finite types.

FormTop/ (Old)

Definitions of formal topology, but computationally relevant parts were hidden in Prop.

Old

  • Prob.v, Prob2.v, Prob3.v : these files are old. They were three different attempts to encode probability in Coq. In Prob.v and Prob2.v, I was hoping to base everything off of the Cantor space, where everything is naturally sample-able.