Paper on reflection

David McAllester (dam@ai.mit.edu)
Thu, 27 Apr 95 10:20:23 EDT

I have looked briefly at your paper on reflection. I think it is a
nice survey and makes a good case for the "LCF suffices" position. I
would define this position as stating that a verification system
should only accept a theorem when a full proof in the base logic has
been generated and verified. This does seem, to me, in opposition to
the "black-box" position which states that for practical
reasons one needs a general way to extend the prover with formally verified
programs whose outputs then do not require further verification.

In human mathematics black boxes are becoming more common. The four
color theorem is the best known example. Here we prove that if a
certain program terminates and prints t then the theorem is true. We
then run the program see that it prints t and believe the theorem.
The proof (the trace of the computation) is too long for humans to
check with confidence.

It seems that black-box theorems, such as the four color theorem,
would be a good test of the "LCF suffices" conjecture. Do you know if anyone
has tried to use a foundational verifier such as HOL to verify any such
black-box theorem?

David