Lets work with some specific way of measuring length of formulas of first order logic theories. Is it possible to have a consistent effectively generated first order theory [i.e. recursively enumerable], that can interpret PA, that is complete for all sentences of its language below a specified length? Of course here we mean sentences that has NO defined symbols in them.
1 Answers
As long as our language is finite, all reasonable notions of length I can think of have the property that there are only finitely many sentences of a given length. This means the answer to the question is trivially yes (e.g. let $PA[n]$ consist of $PA$ + all true sentences of length $<n$).
- It is of course true that the sequence of theories $\langle PA[n]\rangle_{n\in\mathbb{N}}$ isn't effective, but each specific theory $PA[n]$ is.
(And passing to an infinite language doesn't change anything: only finitely much of that language will be necessary to interpret $PA$, so just look at a theory trivializes the remaining symbols; for example set all function symbols to be projection onto the first coordinate, all constants to the element corresponding to $0$, and all relations to the emptyset.)
What if we demand a bit more than mere effectivity - namely, that we have an effective sequence of effective theories with the desired property? That is, we have a computable function $f$ such that for each $i$, $f(i)$ is an index for a consistent computable extension of PA which decides all statements of length $\le i$.
(I'm talking about extensions of PA rather than theories interpreting PA; this doesn't change anything and results in a simpler picture.)
This requirement is strong enough to prevent the silly argument above; in fact, now the answer to the question becomes no, there is no such sequence of theories. The proof isn't as simple as taking the union of the theories involved (since there's no requirement that our theories be jointly consistent, they simply have to be individually consistent). Instead, we use the diagonal lemma as follows.
Let $\langle p_i\rangle_{i\in\mathbb{N}}$ be some fixed effective enumeration of the sentences of arithmetic such that each $p_i$ has length $<i$. Suppose $f$ is as above. Let $q_i$ be $p_i$ if the $f(i)$th theory proves $p_i$ and $\neg p_i$ if the $f(i)$th theory proves $\neg p_i$. Note that the sequence $\langle q_i\rangle_{i\in\mathbb{N}}$ is computable since the function $f$ is computable and the $f(i)$th theory decides $p_i$. Moreover, we know that $PA\cup\{q_i\}$ is consistent for each $i$ (since the $f(i)$th theory extends $PA$ and proves $q_i$).
But from this we get a contradiction. By the diagonal theorem there is an $i\in\mathbb{N}$ such that $p_i$ is equivalent over $PA$ to "$q_i$ is $\neg p_i$" (appropriately expressed). But now we have a problem - since $PA$ is $\Sigma_1$-complete, it can compute each $q_i$ correctly, we can argue:
Suppose $q_i=p_i$. Then $PA$ proves "$q_i$ is not $\neg p_i$," hence $PA$ proves $\neg p_i$. So $p_i$ is not consistent with $PA$, but this contradicts the requirement that $q_i$ is consistent with $PA$.
Suppose $q_i=\neg p_i$. Then $PA$ proves "$q_i$ is $\neg p_i$," hence $PA$ proves $p_i$. So $\neg p_i$ is not consistent with $PA$, but this contradicts the requirement that $q_i$ is consistent with $PA$.
- 245,398
-
How can I know which sentences of length $< n$ are true? – Zuhair Dec 06 '19 at 18:31
-
You don't, but any finite set is recursive. The PAns arent uniformly effective but each is effective individually. – Noah Schweber Dec 06 '19 at 19:02
-
And of course uniform effectivity is impossible since from such a sequence we could effectively choose for each p an element of $ {p, \neg p}$ which is consistent with PA, so that's not a fix. There's no way I see to escape this dichotomy without introducing a very unnatural notion of length. – Noah Schweber Dec 06 '19 at 19:05
-
a small sentence like Fermat's last theorem is something we don't know if it is a true sentence of arithmetic or not? This shows that even for a small length $n$ we don't know which sentences are true. So even if there is in principle an effective theory along the trivial lines you've mentioned, the problem is even for some particular length $n$ still we don't know which sentences of length lower than $n$ are true or not, so there is an effective theory hiding there but we don't know it. – Zuhair Dec 06 '19 at 20:00
-
One of the most important requirements of an effective theory is that we must know which sentences in its language are axioms. The theory "$PA$ + all TRUE sentences of length less than 1000 characters" is not an effective theory, because we don't have effective way to tell which are those sentences, so this is not a satisfactory definition of $PA[1000]$. Until we have such a clear description that single out the true sentence of length say less than 1000 characters, we can't say we have an effective theory just by stipulating it by brute force as such [in the way you did]. – Zuhair Dec 06 '19 at 20:09
-
1@Zuhair "The theory " + all TRUE sentences of length less than 1000 characters" is not an effective theory" I'm sorry, but that's simply incorrect. "Effective theory" is a technical term with a specific meaning (identical to "computable theory" or "recursive theory"), and that theory is indeed effective. It sounds like what you actually want is a uniformly effective sequence of theories, but that's not what you asked; if you're going to use a technical term you can't redefine it to suit your purposes. – Noah Schweber Dec 06 '19 at 23:23
-
@Zuhair I've modified my answer to address the question I think you intended to ask, but it's important to note that $PA[n]$ is indeed an effective theory for each $n$ - that's just what the term means, whether it ought to or no. – Noah Schweber Dec 06 '19 at 23:34
-
Effective axiomatization means that there is a computer program that, in principle, could enumerate all the theorems of the system without listing any statements that are not theorems. For that to occur we need to know explicitly the axioms of the theory. Now just saying PA + all true arithmetical sentences of length < n. [for some particular n] It is not clear what "true" here exactly mean. How can we know if an arithmetical sentence <n length is true or not? If we don't have an explicit description of that, then we cannot have any enumeration, because we don't know the set to be enumerated. – Zuhair Dec 07 '19 at 10:00
-
I understand that in a potential sense one must be able to list all true arithmetical sentences of length < n because there are only finitely many of them, and so they must biject to some initial segment of N, and so can be listed, and moreover the listing function can be definable! So in principle its describable as a scheme added to PA of some finite length. The problem is that for one to know whether an arithmetical sentence is true or not, you need to do that in a complete theory which first order logic cannot define! So you will end up not having an effective system brutally that way. – Zuhair Dec 07 '19 at 10:16
-
A very important feature of an effective theory is that you must be able to tell from the exposition of the theory which sentences in the language of that theory are axioms and which sentence are not! If you cannot tell, then you don't have an effective theory, because it won't be clearly defined theory as simple as such. And to me just saying "all true arithmetic sentences of length < n" to me this is not a clear cut description from which I can know whether a sentence is an axiom of PA[n] or not. Accordingly you cannot have the enumeration required in the definition of effective theory. – Zuhair Dec 07 '19 at 10:24
-
if we don't know the axioms, we don't know the theorems, and so we cannot enumerate what we don't know. For almost all of finite lengths, even though the amount of true arithmetical sentences of each length is finite, yet brute listing of most of these amounts is not possible in real life. So you need a write-able expression that can isolate those sentences. And that's what the question is about – Zuhair Dec 07 '19 at 11:46
-
@Zuhair Again, that's just not what that term "effective theory" means. You might argue that it should mean something else, but the fact is that foreach $n$ the theory $PA[n]$ is an effective theory (just like how for each $n$ the set $A_n={x: x=0$ and the $n$th sentence in the language of $PA$ is true$}$ is a computable set even though for most $n$ we don't know whether $A_n$ is ${0}$ or $\emptyset$). – Noah Schweber Dec 21 '19 at 18:04