I just gave a talk at “Semantics of proofs and certified mathematics”. I spoke about a new proof checker Chris Stone and I are working on. The interesting feature is that it has both kinds of equality, the “paths” and the “strict” ones. It is based on a homotopy type system proposed by Vladimir Voevodsky. The slides contain talk notes and explain why it is “Brazilian”.
Download slides: brazilian-type-checking.pdf
GitHub repository: https://github.com/andrejbauer/tt
Abstract: Proof assistants verify that inputs are correct up to judgmental equality. Proofs are easier and smaller if equalities without computational content are verified by an oracle, because proof terms for these equations can be omitted. In order to keep judgmental equality decidable, though, typical proof assistants use a limited definition implemented by a fixed equivalence algorithm. While other equalities can be expressed using propositional identity types and explicit equality proofs and coercions, in some situations these create prohibitive levels of overhead in the proof.
Voevodsky has proposed a type theory with two identity types, one propositional and one judgmental. This lets us hypothesize new judgmental equalities for use during type checking, but generally renders the equational theory undecidable without help from the user.
Rather than reimpose the full overhead of term-level coercions for judgmental equality, we propose algebraic effect handlers as a general mechanism to provide local extensions to the proof assistant’s algorithms. As a special case, we retain a simple form of handlers even in the final proof terms, small proof-specific hints that extend the trusted verifier in sound ways.
On December 6th 2011 I gave a talk about homotopy equivalences in the context of homotopy type theory at our seminar for foundations of mathematics and theoretical computer science. I discuss the differences and relations between isomorphism (in the sense of type theory), an adjoint equivalence, and a homotopy equivalence. Even though the talk itself was not super-well prepared, I hope the recording will be interesting to some people. I was going fairly slowly, so it should be possible to follow the talk. I apologize for such a long video, but I really did not see how to chop it up into smaller pieces. Also, I need to figure out why I cannot fast forward the video beyond what has been downloaded.
A talk given at “Mathematics, Algorithms and Proofs 2011” at the Lorentz Center in Leiden, the Netherlands. I explain how to use computational effects to speed up Martin Escardo’s impossible functionals.
A talk given at “Computation with Infinite Data: Logical and Topological Foundations” Dagstuhl seminar 11411. I describe a realizability model based on infinite-time Turing machines in which it is possible to embed the Baire space (infinite sequences of numbers) into the space of numbers.
Also see the post Constructive gem: an injection from Baire space to natural numbers for written notes on this topic.
[UPDATE 2012-03-08: since this post was written eff has changed considerably. For updated information, please visit the eff page.]
I just returned from Paris where I was visiting the INRIA ?r² team. It was a great visit, everyone was very hospitable, the food was great, and the weather was nice. I spoke at their seminar where I presented a new programming language eff which is based on the idea that computational effects are algebras. The language has been designed and implemented jointly by Matija Pretnar and myself. Eff is far from being finished, but I think it is ready to be shown to the world. What follows is an extended transcript of the talk I gave in Paris. It is divided into two posts. The present one reviews the basic theory of algebras for a signature and how they are related to computational effects. The impatient readers can skip ahead to the second part, which is about the programming language.
A side remark: I have updated the blog to WordPress to 3.0 and switched to MathJax for displaying mathematics. Now I need to go through 70 old posts and convert the old ASCIIMathML notation to MathJax, as well as fix characters which got garbled during the update. Oh well, it is an investment for the future.
Continue reading Programming with effects I: Theory
These are the slides and the extended abstract from my MSFP 2008 talk. Apparently, I forgot to publish them online. There is a discussion on the Agda mailing list to which the talk is somewhat relevant, so I am publishing now.
Abstract: Realizability is an interpretation of intuitionistic logic which subsumes the Curry-Howard interpretation of propositions as types, because it allows the realizers to use computational effects such as non-termination, store and exceptions. Therefore, we can use realizability as a framework for program development and extraction which allows any style of programming, not just the purely functional one that is supported by the Curry-Howard correspondence. In joint work with Christopher A. Stone we developed RZ, a tool which uses realizability to translate specifications written in constructive logic into interface code annotated with logical assertions. RZ does not extract code from proofs, but allows any implementation method, from handwritten code to code extracted from proofs by other tools. In our experience, RZ is useful for specification of non-trivial theories. While the use of computational effects does improve efficiency it also makes it difficult to reason about programs and prove their correctness. We demonstrate this fact by considering non-purely functional realizers for a Brouwerian continuity principle.
Download: msfp2008-slides.pdf, msfp2008-abstract.pdf
Two versions of this talk were given at Computability and complexity in analysis 2008 and at Mathematics, Algorithms and Proofs 2008.
Joint work with Paul Taylor.
Abstract: Cauchy’s construction of reals as sequences of rational approximations is the theoretical basis for a number of implementations of exact real numbers, while Dedekind’s construction of reals as cuts has inspired fewer useful computational ideas. Nevertheless, we can see the computational content of Dedekind reals by constructing them within Abstract Stone Duality (ASD), a computationally meaningful calculus for topology. This provides the theoretical background for a novel way of computing with real numbers in the style of logic programming. Real numbers are defined in terms of (lower and upper) Dedekind cuts, while programs are expressed as statements about real numbers in the language of ASD. By adapting Newton’s method to interval arithmetic we can make the computations as efficient as those based on Cauchy reals.
Slides: slides-map2008.pdf (obsolete version: slides-cca2008.pdf)
Extended abstract: abstract-cca2098.pdf
With Iztok Kavkler.
Abstract: The interval domain was proposed by Dana Scott as a domain-theoretic model for real numbers. It is a successful theoretical idea which also inspired a number of computational models for real numbers. However, current state-of-the-art implementations of real numbers, e.g., Mueller’s iRRAM and Lambov’s RealLib, do not seem to be based on the interval domain. In fact, their authors have observed that domain-theoretic concepts such as monotonicity of functions hinder efficiency of computation.
I will review the data structures and algorithms that are used in modern implementations of exact real arithmetic. They provide important insights, but some questions remain about what theoretical models support them, and how we can show them to be correct. It turns out that the correctness is not always clear, and that the good old interval domain still has a few tricks to offer.
Download slides: domains8-slides.pdf
With Davorin LeÅ¡nik.
Abstract: We investigate the relationship between constructive theory of metric spaces and synthetic topology. Connections between these are established by requiring a relationship to exist between the intrinsic and the metric topology of a space. We propose a non-classical axiom which has several desirable consequences, e.g., that all maps between separable metric spaces are continuous in the sense of metrics, and that, up to topological equivalence, a set can be equipped with at most one metric which makes it complete and separable.
Presented at: 3rd Workshop on Formal Topology
Download slides: 3wft.pdf