0
$\begingroup$

We have numerous logical systems. For example PA is a (sort of) formal definition of number theory. PA can (I guess) be seen as a fragment of ZFC, which makes it weaker than ZFC. Further there seem to be numerous fragments of PA which are, of course, strictly weaker than PA. This shows there exists an ordering between the systems wrt. to their expressiveness.

In computation theory we usually deploy a Turing-complete model and restrict that to get less expressive "fragements". For example, we can put time constraints and deal with total polynomial time Turing machines, or we can define a programming language that quarantees halting of the programs written with it and, hence, it falls short in computational power from Turing machines. This shows we have some kind of ordering here, too.

Now, these two concepts seem to be closely related but I don't understand how. Is it possible to, say, use PA to define a programming language that would be a fragment of Turing-complete language? Or could we use Turing-complete model to define a formal logic that would be as expressive as one could ever be (I guess not)? Or are logic and computation after all incomparable but they just happen to interwind?

I'm aware of Curry-Howard correspondence, but that seems to just deepen my confusion on the matter. I'm also aware that some logics coincide with computational complexity classes and I have studied some finite model theory, but that hasn't really helped to understand the big picture.

I know the question goes a bit vague and I would be perfectly happy with pointers to literature that would explain what is going on.

  • 0
    A derivation in a formal system is a computation: since the 1950s we have softwares that can produce derivations automatically. – 2017-02-01
  • 0
    @MauroALLEGRANZA Yes, true. I also realize the relation between incompleteness theorem and halting problem, but the big picture gives some hard time to understand. – 2017-02-01
  • 1
    One (technically "convoluted" but conceptually quite straightforward) of the way to show the "reducibility" is the technique called [Arithmetization of Syntax](http://www.michaelbeeson.com/teaching/StanfordLogic/Lecture12Slides.pdf) due to Gödel. See also [Gödel's Incompleteness Theorems](https://plato.stanford.edu/entries/goedel-incompleteness/#AriForLan). – 2017-02-01
  • 1
    Useful references : George S. Boolos, John P. Burgess, Richard C. Jeffrey, [Computability & Logic](https://books.google.it/books?id=8ABhQgAACAAJ) (5th ed 2007), Raymond Smullyan, [Godel's Incompleteness Theorems](https://books.google.it/books?id=04zalcCdKZsC&printsec=frontcover) (1992), Vladimir A. Uspenski, [Gödel's incompleteness theorem](https://books.google.it/books?id=60pZAAAAYAAJ) (Engl.transl., 1987). – 2017-02-01
  • 1
    Maybe usefule (parts of) : Marvin Minsky, [Computation : Finite and Infinite Machines](https://books.google.it/books?id=CURDAAAAIAAJ) (1967), Richard Feynman, [Lectures on Computation](https://books.google.it/books?id=NCJQq1_ZKgEC) (1966) and Cristopher Moore & Stephan Mertens, [The Nature of Computation](https://books.google.it/books?id=z4zMiZyAE1kC&pg=PR7) (2011). – 2017-02-01

1 Answers 1

1

One way find a link between computation and formal theories is to look at so-called provably computable functions (provably recursive functions). These are the total computable functions which can be proved to be total in a particular theory. One of the main motivating problems in proof theory has been to characterize this set of functions for various theories such as Peano Arithmetic.

This does give us a hierarchy of subclasses of the total computable functions. It is known that not every total computable function is provable total in PA; there are precise characterizations in proof theory of the set of provably total functions of PA. On the other hand, if we weaken the induction scheme in PA to only include induction for $\Sigma^0_1$ formulas, the resulting has has exactly the primitive recursive functions as its set of provably computable functions.

There are theories much weaker than PA, known as theories of bounded arithmetic, whose provably computable functions are related to computational complexity classes, although not in a way I can summarize easily with my limited knowledge of bounded arithmetics.

Unfortunately, I don't think there is any undergraduate-level literature on these subjects, which means that the only sources are graduate and professional level proof theory and computer science writing. That makes it a difficult area to learn on your own, if you don't have a deep background in logic. The Handbook of Proof Theory has an article "Hierarchies of Provably Recursive Functions" by Stanley S. Wainer could be a starting point.

  • 0
    I'm fine with graduate level material. This has developed to a half-serious hobby for me that I have pushed for years now. I have self studied for long time and got MSc in math/logic (finite model theory) and I have a half-written dissertation in subrecursive hierarchies (with all course work completed for PhD degree), but this progresses slowly as I get lost on these side tracks continuously, and I work it by myself on my spare time :) – 2017-02-01
  • 0
    It sounds like you are in the fortunate position to be able to approach the literature, in that case. I usually include a disclaimer because, on a general site, many people ask questions for which the answer is known but the literature is too difficult to be a useful resource. If other people read the answer later, they may wonder if there is a more accessible summary, but in the case of proof theory I don't think there is. You may also be interested in Goedel's "System T", which is used both in proof theory and theoretical computer science. – 2017-02-01
  • 0
    As it happens I'm aware of the concept of provably total function, but now that I think of it... I find I must think this a bit further from what I thought of earlier. (Somehow I feel there is more to the subject that is to be found.) – 2017-02-01
  • 0
    I have come across Gödel's Dialectica interpretation. I think one could use it to redefine the fast-growing hierarchy in (somewhat) more natural way with the concept of higher order recursion Gödel defined, but it was too hard for me to progress. – 2017-02-01
  • 0
    @FFF: Incidentally, I came up with a rather similar idea recently after writing a program for http://chat.stackexchange.com/transcript/message/35359610#35359610, namely http://chat.stackexchange.com/transcript/message/35530869#35530869, where I used a lemma that is naturally stated in third-order arithmetic. I don't know much but it feels like we can do a lot with predicative higher-order arithmetic (predicative meaning that the defining formulae for a $k$-th order collection only quantifies over $k$-th order collections). Carl, can I ask, do you know the proof-theoretic strength of this? – 2017-02-19
  • 0
    @Carl Mummert, I continued learning about strong fragments of PA. My ultimate goal was to find new tools to built subrecursive hierarchies, but I came up (half accidentally) with a characterization of fragments $I\Sigma_n$ of PA. Basically I have a proof that each $I\Sigma_n$ is a time complexity class (every $f\in I\Sigma_n$ is computable in time $g\in I\Sigma_n$ and vice versa). Do you know if this is a known result? – 2018-01-17
  • 0
    There is a lot known about computability and proof theory of fragments of PA. When you write "$f \in I\Sigma_n$" do you mean that $f$ is a computable function $\phi_e$ so that $I\Sigma_n$ proves that $f$ is total? I am not sure exactly what that notation means. @zzz – 2018-01-17
  • 0
    (Sorry about the ambiguity, one shouldn't type in hurry..) I mean that if $f$ can be proved total in $I\Sigma_n$ there is $g$ provably total in the same fragment such that $f$ is computable in time $g$, and if $f$ can be computed in time $g$ for $g$ provable total in $I\Sigma_n$ then $f$ is provable total in $I\Sigma_n$. – 2018-01-17
  • 0
    If $T$ is any theory of arithmetic that proves $(\forall n)(\exists s)(\exists z) [\phi_{e,s}(n)\downarrow = z]$, then this simultaneously determines a function $f(n) = z$ and the time function $g(n) = s$, and so if the theory proves either of those functions is total then it proves the other one is total. I am not sure that this is what you mean, though? – 2018-01-17
  • 0
    Yes, this seems to be what I was after to. Thank you. What I had initially is that the elementary recursive functions $\mathcal{E}$ can be extended with a total computable $f$ (resulting to $\mathcal{E}(f)$) such that the extension is a time complexity class, i.e. if $h\in\mathcal{E}(f)$ then for some $g\in\mathcal{E}(f)$ we have $h\in\mathcal{O}(g)$ and the opposite, i.e. if $h\in\mathcal{O}(g)$ with $g\in\mathcal{E}(f)$ then $h\in\mathcal{E}(f)$. And this extends to fragments of PA. – 2018-01-18