1
$\begingroup$

My motivation to this question is to quantify whether or not Gentzen's consistency proof of Peano arithmetic could be considered acceptable under a constructive framework.

I admit that in some sense this is not a well-defined question, as "constructive" can have a variety of meanings, so for the purposes of this question, take "constructive" to mean that we are either working under a common constructive framework (IZF, IZF + CT, MLTT, etc...), or are simply going from possibly informal constructive principles, such as Brouwer's intuitionism that at least some constructivists would find acceptable.

  • 0
    Please include your question in the body of the post.2017-01-12
  • 0
    What do you mean by "finite rooted trees"?2017-01-12
  • 0
    @Rob https://en.wikipedia.org/wiki/Tree_%28graph_theory%292017-01-13
  • 0
    @MattSamuel: thanks for your input. If the OP's "finite rooted trees" are just finite trees in the usual mathematical sense, then they are obviously constructively well-orderable. Perhaps the OP would like to comment.2017-01-13
  • 0
    @Rob They're finite *rooted* trees. Rooted trees are addressed in the article. And what is the obvious well ordering?2017-01-13
  • 2
    @Matt: they're finite data structures. Encode them as natural numbers using the usual tricks and you get a gazillion ways of well-ordering them constructively.2017-01-13
  • 0
    In his paper on the Brouwer-Hilbert Controversy, Martin Lof specifically says that the ordinal $\epsilon_0$ (whose well-orderability is equivalent to that of finite rooted trees) is "an ordinal about whose constructive accessibility nobody has any doubt." -- I suppose because of the "usual tricks" that Rob mentions (which now I think I understand the general procedure for).2017-02-13
  • 0
    Actually, for completeness sake, I may attempt to compose my own answer to this question so people can upvote/downvote it to determine if I have properly understood the technique.2017-02-13

1 Answers 1

1

Suppose our underlying logic is such that we do not assume LEM for statements whose truth-value is uncomputable. But suppose that we accept inductive constructions for natural numbers, and we have higher-order types. I shall also write "$x_k$" for "$x(k)$" where $x$ is a sequence and $k$ is a natural number.

Call a type $S$ well-ordered under $<$ iff every sequence from $S$ has a minimum under $<$. $ \def\nn{\mathbb{N}} \def\then{\mathrel{?}} $

We can then construct and prove a well-ordering on finite (rooted) trees (which in ZFC would have order-type $ε_0$) as follows.

First we prove a lemma:

Given any $S$ that is well-ordered under $\le$, the type $T$ of (non-strict) $\le$-decreasing sequences from $S$ is well-ordered under lexicographic order.

Proof:

Take any sequence $f$ from $T$. Namely, $f_0,f_1,f_2,\cdots$ are $\le$-decreasing sequences from $S$.

Let $g = \Big( \nn\ n \mapsto \min\big( \nn\ k \mapsto \forall x\in\nn_{

[By induction, given $f$ and $\le$ on $S$ and $\min$ on $(\nn \to S)$ we can compute each item of $g$.]

Then $g$ is a sequence from $S$, and hence let $m = \min(g)$ and $c \in \nn$ such that $g(c) = m$.

Also $g$ is $\le$-decreasing, and hence $g(n) = m$ for every $n \in \nn_{\ge c}$.

Let $d \in \nn$ such that $f_d(x) = g(x)$ for every $x \in \nn_{\le c}$, which exists by induction.

Then by definition of $g$ and induction, $f_d(x) = g(x)$ for every $x \in \nn_{>c}$, and hence $g = f_d \in T$.

Finally by one more induction, $g \le_{lex} f_k$ for every $k \in \nn$.

I will leave you to fill in the details.

Now define the ordering of finite trees inductively. All trees of depth at most $0$ are equal. Compare two trees of depth $k$ by first sorting their sequences of subtrees in decreasing order (which is valid because their subtrees have depth less than $k$) then comparing them lexicographically. By the lemma and induction, the resulting comparison is a well-ordering on the trees of depth $k$. And compare two trees of different depth by their depth.

We can then easily prove that the resulting comparison is a total-ordering on the finite trees. And then we can prove that it is a well-ordering on the finite trees, since for every sequence $f$ of trees we can construct $g = ( \nn\ n \mapsto n=0 \then f(0) : \min(g(n-1),f(n)) )$ inductively, which then is a decreasing sequence of trees, whose depths are eventually constant at some $k \in \nn$, from which point the sequence has a minimum by the above argument.


Of course, first-order PA is not strong enough to express the above proof, since it implies Con(PA). However, another way in which the well-ordering of $ε_0$ is constructive is that PA can construct the above-defined ordering and also prove each instance of the usage of the lemma in the above proof, and hence can prove that the ordering is a well-ordering on trees of depth at most $k$, for each fixed meta-natural $k$.

  • 0
    Also, the argument can be formalized in ACA. So if you accept finite Turing jumps as having well-defined truth-values, and you accept full induction (the schema for all second-order formulae), then you would accept this argument suitably translated. It's easier to understand in the format I've given, because inductive constructions are a pain to perform in standard classical systems like ACA.2017-03-27