|From:||Gabriel Dos Reis|
|Subject:||Re: [Axiom-developer] Proving Axiom Correct|
|Date:||Wed, 25 Jan 2017 00:18:02 -0800|
He makes a distinction betwen invented and discovered langauges.> There were implementations of C in Lisp. So C shares that formal logic basis, or that it was discovered?According to Wadler, C is an "invented" language, not a "discovered" language.shows the logician's view versus the computer science view of various theories.
watch?v=aeRVdYN6fE8) at around 40 minutes
(begin quote of Wadler)
Every interesting logic has a corresponding language feature:
Natural Deduction (Gentzen) <==> Typed Lambda Calculus (Church)
Type Schemes (Hindley) <==> ML Type System (Milner)
System F (Girard) <==> Polymorphic Lambda Calculus (Reynolds)
Modal Logic (Lewis) <==> Monads (state, exceptions) (Kleisli, Moggi)
Classical-Intuitionistic Embedding (Godel) <==> Continuation Passing (Reynolds)
Linear Logic (Girard) <==> Session Types (Honda)
Languages which were "discovered" (based on theory):
Scheme (Steele and Sussman)
ML (Milner, Gordon, Wadsworth)
Haskell (Hudak, Hughes, Peyton Jones, Wadler)
Erlang (Armstrong, Virding, Williams)
At about minute 43 we hear:
"Not all of these languages are based on lambda calculus ...
but there is a core that is "discovered". Now most of you work
in languages like Java, C++, or Python and those languages I will
claim are not discovered; they are "invented". Looking at them
you can tell they are invented. So this is my invitation to you
to work in languages that are "discovered".
Somebody asked before about "dependent types". It turns out that
when you extend into dependent types you now get languages that
have "for all" and "there exists" that corresponds to a feature
called "dependent types". So this means that you can encode very
sophisticated proofs, pretty much everything you can name as a
program. And so the standard way to get a computer to check a
proof for you and to help you with that proof is to represent that
as a program in typed lambda calculus.
All of these systems are based on a greater or lesser extent on
Automath (de Bruijn)
Type Theory (Martin Lof)
ML/LCF (Milner, Gordon, Wadsworth)
HOL (Gordon, Melham)
Coq (Huet, Coquand)
Epigram (McBride, McKinna)
Axiom is using Coq for its proof platform because Axiom needs
dependent types (e.g. specifying matrix sizes by parameters).
In addition, Coq is capable of generating a program from a
proof and the plan is to reshape the Spad solution to more
closely mirror the proof-generated code. Also, of course, we needto remove any errors in the Spad code found during proofs.
It seems there must be an isomorphism between Coq and Spad.
At the moment it seems that Coq's 'nat' matches Axiom's
NonNegativeInteger. Coq also has a 'Group' type which needs
to be matched with the Axiom type. The idea is to find many
isomorphisms of primitive types which will generate lemmas
that can be used to prove more complex code.Given that Axiom has an abstract algebra scaffold it seems likelythat a reasonable subset can be proven (modulo the fact that thereare arguable differences from the abstract algebra type hierarchy).
Currently Coq is run during the Axiom build to check any proofs
of Spad code that are included. ACL2 is also run during the build
to check any proofs of Lisp code that are included.
I'm taking a course at Carnegie-Mellon this semester using Lean
(a Coq offshoot) in order to try to make some working examples
of proving Spad code correct.On Thu, Jan 12, 2017 at 10:14 PM, Gabriel Dos Reis <address@hidden> wrote:There were implementations of C in Lisp. So C shares that formal logic basis, or that it was discovered?On Wed, Jan 11, 2017 at 8:17 PM, Tim Daly <address@hidden> wrote:______________________________the formal logic basis.just a domain specific language implemented in Lisp then Spad sharesThe point is that Lisp has a formal logic basis and, as Spad is reallywhether your language is discovered or invented".are "invented" with no formal basis (e.g. C). As he says, "you can tell"discovered" based on fundamental logic principles (e.g.Lisp) and othersHe makes the interesting point late in the talk that some languages areTypes", a very entertaining talk, is here:I'm making progress on proving Axiom correct both at the Spad level andthe Lisp level. One interesting talk by Phillip Wadler on "Propositions as
Axiom-developer mailing list
|[Prev in Thread]||Current Thread||[Next in Thread]|