Find an infinite sequence of statements so that truthness of all sequence elements is unprovable

99 Views Asked by At

Let us consider a function $f: \mathbb N \to \left\{0, 1\right\}$ with finite description length (i.e. describable by a finite length program for a Turing machine) satisfying the following conditions:

  1. $f(n)$ is computable in finite time for any $n,$ that means that it is possible to prove that the program for a Turing machine that calculates $f(n)$ always halts;
  2. It is impossible to compute in finite time $\prod\limits_{n=1}^{\infty} f(n),$ i.e. there is no program for a Turing machine that in finite time checks the fact that $\forall n: f(n) = 1$ and outputs $1$ in this case and $0$ otherwise.

The questions are:

  1. Is it possible to prove that such function exists or does not exist?
  2. If it is possible, are there some concrete examples?

Here are some my thoughts.

One possible candidate for such function is $f(n) = I[a_n = 1]$ where $\left\{a_n\right\}$ is Collatz sequence and $I[\cdot]$ is the indicator function. However for me it is not clear is it possible to prove Collatz conjecture to be unprovable or we couldn't even prove that unprovability.

I would appreciate any reading recommendations on this topic.

1

There are 1 best solutions below

4
On BEST ANSWER

There are some problems with your function. First, notice that in (1) it's not actually the same to say "$f(n)$ is computable in finite time for every $n$" and "it is provable that the Turing machine computing $f$ halts on every input". A good counterexample is the Turing machine which, regardless of its input, searches for a proof of a contradiction from Peano Arithmetic; this Turing machine always fails to halt (provided PA is consistent) but that fact can't be proven in Peano Arithmetic.

In (2), $\prod_{n=1}^{\infty}f(n)$ is a single number (either $1$ or $0$). A single number is always computable, in this case either by the Turing machine that always outputs $0$ or by the machine that always outputs $1$. Now, we may not be able to decide which one is the right one, but that's a provability question rather than a computability one. In general, in order for it to make sense to say something is noncomputable, you need it to be infinite - usually, an infinite sequence.

On the other hand, you can have this: a sequence of binary functions $f_i$ for $i = 0,1,2,\ldots$ so that each $f_i$ is computable, but the sequence $p_i = \prod_{n=1}^{\infty}f_i(n)$ is not. For a natural example, let $f_i(n) = 0$ if the $i$th Turing machine halts before stage $n$ and $1$ otherwise. Then $p_i$ is $0$ if the $i$th Turing machine halts or $1$ if it doesn't; that makes computing the $p_i$ equivalent to computing the Halting Problem.