12:08 AM
@Threnody Yes, floor(r·2^(k+1)) mod 2 is the k-th binary digit. I was lazy to check hahaha.. I don't get your first question about Internal IVT; the usual formulation does assume f(0)<0<f(1). Your second question is a good one; it fails. Consider f = ( ℝ x ↦ x<√2 ? −1 : 1 ). Then there is no rational x such that ∀p∈ℚ ( ( p<x ⇒ f(p)≤0 ) ∧ ( p>x ⇒ f(p)≥0 ) ). It also disproves your claim that you can extend every function from ℚ→ℝ to make it continuous.
In any case, if your claim were true, then I shouldn't be giving you the "fun exercise" because it would have a totally unnecessary condition "f encodes a continuous function".
As for quantifying over ℝ→ℝ, of course you can't. That's why half my explanation was about encoding and decoding because that is the only way to quantify over ℚ→ℝ. In particular:
7 hours ago, by user21820
Back to encoding continuous functions. By the earlier described encoding of sequences of reals, the quantification "∀f∈(ℚ→ℝ)" is essentially just a "∀f∈ℝ" with suitable decoding of f to extract the values of the decoded function on ℚ. So we are not going beyond the capability of the theory T.
For example, smallest prime factor is definable over PA, by the predicate Q(k,p) ≡ ( k≤1 ∧ p=0 ) ∨ ∃x∈ℕ ( p>1 ∧ k=p·x ∧ ∀q ( 1<q<p ⇒ ∀y∈ℕ ( k≠q·y ) ) ).
PA proves ∀k∈ℕ ∃!p∈ℕ ( Q(k,p) ), so by (2) of the linked post we can add a new function symbol pfactor to PA and the axiom ∀k,p∈ℕ ( pfactor(k)=p ⇔ Q(k,p) ), and the resulting theory proves exactly the same sentences over the language of PA (i.e. that do not use the new symbol pfactor).
Oops I blundered in defining Q. It should be Q(k,p) ≡ ( k≤1 ∧ p=0 ) ∨ ( k>1 ∧ ∃x∈ℕ ( p>1 ∧ k=p·x ∧ ∀q ( 1<q<p ⇒ ∀y∈ℕ ( k≠q·y ) ) ) ).
And the point of defining new symbols is so that we don't have an utterly gigantic headache when doing encoding and decoding. For instance, after defining pfactor, we can state theorems like ∀k∈ℕ ( k>1 ⇒ pfactor(pfactor(k)) = pfactor(k) ).
12:38 AM
Now for the encoding/decoding issue for ℕ→ℝ. I am essentially claiming that there a bijection from ℝ to (ℕ→ℝ) via the described decoding, and we can define a new function symbol extract with signature ⟨ℝ,ℕ⟩→ℝ that agrees with that bijection! That is, for every actual real r and actual natural k, we have decode(r)(k) equal to ( extract(f,k) when ℕ,ℝ are interpreted as the actual naturals and reals ).
@Threnody Finally, about "I gave T too much power", I wanted to say that T does not prove Internal IVT, but I could not prove that T does not prove Internal IVT after thinking for a while, so I gave up and used Internal Sup instead. My goal was to show that what appears to be a good combination of PA and the FOL theory of the reals is not actually enough for basic real analysis, so I needed to find some real analysis theorem that could not be proven by T.
By the way, the power of these kind of theories arise from the axioms that assert existence of reals. The theory of ordered fields doesn't force much to exist except (a copy of) the rationals. But the FOL theory of the reals has some existential axioms, which force closure under algebraic extension and hence force the existence of all algebraic reals. Internal Sup provides a lot more power than that; it forces closure under sup of internal sequences (i.e. sequences coded by some element of ℝ).
Oops, I forgot to add comprehension axioms to SRA. Lol I was more careful in my post on the main site; I used a schema because I was lazy to add comprehension axioms (contrary to what I said earlier in chat). Let me rectify that.
> SRA (version 1) (following the post on Main): PA for ℕ plus FOL theory for ℝ plus ( ∃x∈ℝ ( f≤x ) ⇒ ∃m∈ℝ ( f≤m ∧ ∀y∈ℝ ( f≤y ⇒ m≤y ) ) for every f with signature ℕ→ℝ that is definable over SRA with parameters.
> SRA (version 2) (what I want to describe here): PA for ℕ plus FOL theory for ℝ plus Internal Sup plus ( ∃f∈(ℕ→ℚ) ∀k∈ℕ ∀x∈ℚ ( f(k)=x ⇔ R(k,x) ) ) for every predicate R with signature ⟨ℕ,ℚ⟩→bool.
I forgot because usually in reverse mathematics we have comprehension axioms for subsets of ℕ (rather than for real numbers), and I thought I could get away with just Internal Sup, but I forgot that I still didn't have a way to construct sequences to start with...
Anyway, here I shall only use version 2. PA for ℕ plus FOL theory for ℝ is the bare minimum that we obviously want to have. The comprehension axiom schema at the end is completely reasonable as well, because it just says that we can construct rational sequences given a defining formula.
So the powerful tool here is Internal Sup. For example, let H = ( ℕ k ↦ ( the k-th TM halts on input 0 ) ? 1 : 0 ). H is known as the halting oracle. Note that H is definable over PA. Then SRA can construct an element of ℝ encoding H: Let f = ( ℕ k ↦ Sum { H(i)/2^i : i∈{1..k} } ). Then f is a definable rational sequence over SRA so we can construct it in SRA (i.e. there is an element of ℝ that encodes it). And Internal Supremum says it has a supremum, which we know is uncomputable like H.
6 hours later…
7:50 AM
8:07 AM
1 hour later…
9:18 AM
Because being able to add newly defined function-symbols given a defining predicate for which you can only prove existence will cause ZF to become ZF plus GC (global choice), and of course ZF+GC proves AC.
@Threnody Correct. Always think of TMs as programs in some idealized programming language. You can list all valid programs one by one in order of length-lexicographic order. This is a computably enumerable list, and that is all that is important. The k-th program is defined as the k-th program in that list.
10:14 AM
@user21820 Does this have anything to do with 'well defined' as I usually see it in textbooks? i.e. x = y => f(x) = f(y)
2 hours later…
12:46 PM
@Threnody No. Textbooks that use the phrase "well defined" and lead you to write what you just did are lousy.
Given any objects x,y such that x=y, we always have f(x) = f(y) for any function f whatsoever. If you don't then your proof must have gone bonkers somewhere.
For example of a bonkers poof: Let f : ℚ→ℕ such that f(a/b) = a for every a,b∈ℤ such that b>0. (WRONG! You cannot construct such an f. Nothing else to say here.)
For example of a correct proof: Suppose we have the naturals ℕ but not integers and we want to construct integers. Let ~ be a relation on ℕ^2 defined by ⟨a,b⟩ ~ ⟨c,d⟩ iff a+d = b+c. Then ... Thus ~ is an equivalence relation on ℕ^2. Let ℤ = { { p : p∈ℕ^2 ∧ p~q } : q∈ℕ^2 }. Let ⊕ be a function on ℤ^2 such that p+q = { ⟨a+c,b+d⟩ : a,b,c,d∈ℕ ∧ ⟨a,b⟩~p ∧ ⟨c,d⟩~q }, for every p,q∈ℤ. Note that I didn't say ⊕ : ℤ^2→ℤ because we haven't proven it yet! Then ... Thus ∀p,q∈ℤ ( p⊕q∈ℤ ). Furthermore ...
Observe carefully that in a proper rigorous proof you do not ever need to use the term "well-defined", because every object you construct in a true proof is already well-defined (because the soundness of an FOL deductive system guarantees it).
1:57 PM
@Threnody It is possible to rigorously construct a relation first and prove it encodes a function later. But if a textbook does not do that either, then it is almost surely non-rigorous or simply wrong. You can ask in Basic Mathematics if you see any such example and you want my comment on it.
Actually it may be of interest to you to note that you can introspect yourself to be able to understand what I say concerning "well-defined" much faster than I think most students can, simply because you have a better understanding of formal deduction.
Some students when told that you cannot define something just by writing down what you want it to be like because it is not allowed, ask, "why not?" and are not satisfied with answers like "because it may not exist and so we do not allow you to do such a thing". In some sense, many people have the erroneous notion that what they can describe must exist...
I see... I've never seen Z constructed this way... well to be fair I've never seen Z constructed in any way.
3 hours later…
4:56 PM
3
A: How to universalize $\text{Prov}(\ulcorner y < K(x)\urcorner) \to y < K(x)$ in a paper of Kikuchi
It seems that the text is using the lemma (arithmetized $Σ_1$-completeness of PA) for $Σ_1$-formulae rather than just sentences. Originally, I had thought that the generalized version could be easily proven from the specialized one, but I made a careless mistake. Now I believe that it cannot be p...
2 hours later…
« first day (1481 days earlier) ← previous day next day → last day (1330 days later) »