Proving $A_{TM}$ is mapping reducible to certain language

Question

I've been asked to prove that the language $A_{TM} = \{ \langle M,w\rangle \mid M$ is a TM that accepts $w\}$ is mapping reducible to the language $LOOP-ONE = \{\langle M \rangle \mid M$ is a Turing machine that doesn't halt on exactly one input$\}$.

I came up with the following idea :

F = "On input <M,w>:

1. Construct the following machine M':
   M' = "On input x:
   1.If $x=w \cdot 1$, enter a loop. (assuming $1 \in \Sigma$ )
   2.Run M on w.
   3.If M accepts, accept. otherwise, enter a loop."
2. Output <M'>."

As for the correctness of the reduction above we can see that :
If $<M,w> \in A_{TM} \implies M$ accepts $w$ therefore by the construction above $M'$ accepts any input $x$ (besides $w\cdot1$) therefore in particular $M'$ halts on any input other than $w\cdot1$.
On the other hand,
If $<M,w> \notin A_{TM} \implies M$ doesn't accepts $w$ therefore $M$ either loops on $w$ or rejects $w$.
If $M$ loops on $w$ then $<M'> \notin LOOP-ONE$ as $M'$ doesn't halts on $w$ and on $w\cdot1$.
If $M$ rejects $w$ then $M'$ will loop on any input $x$ and therefore $<M'> \notin LOOP-ONE$.

I was asked to make sure that the function that the TM F is computing is indeed computable, and I'm struggling to see what it practically means. What are the things that make a mapping reduction not computable (besides maybe the fact that the Turing machine F may not halt on some input, which is not the case here) ?

Answer 1

For a reduction $f$ to be computable, there must be a TM that for all inputs $y$ halts with $f(y)$ written on its tape. The reduction you suggested is indeed computable. Here, your input $y$ is assumed to be of the form $\langle M, w \rangle$, then the reduction outputs a description of a machine $\langle M'\rangle$. The crux of the idea here is to notice that the reduction $f$ (or the machine that tries to implement it) does not really simulate the run of $M$ on $w$, and does not simulate the run of $M'$ on some input. It just outputs the description of $M'$ and then halts. So you only need to convince the reader that there is a TM $T_{f}$ that can output the description of $\langle M'\rangle$ based only on the description of $\langle M, w\rangle$, and then halt. This is doable as you can take the description of $M'$ to be a description of a universal TM that has the description of $M$ and $w$ encoded in it (for example in its state-space). Also $M'$ has a state $q_{loop}$ that forces it to inter a loop. Then, on input $x$, what $M'$ should be able to do is:

• Apply a simple check of the input's format: it is easy to describe a machine that does that and halt.

• Simulate the run of $M$ on $w$: we can describe also this component (or part) of the machine $M'$ similarly to the description of a universal TM.

• Enter a loop: we can encode the transition function of $M'$ to enter $q_{loop}$ in some scenarios.

So the main thing here is to note that the description of $M'$ can be obtained by manipulating the description of $M$ and $w$ and putting that together with a description of universal TM that does simple checks. So you can think of $M$ ia a python code, and $w$ is its input. Then, you (the reduction), given this code and input, have to output a python code $M'$ that behaves in a way that depends on $M$ and $w$. That is doable, as long as you (the reduction) sew the codes together without really running them, and so the risk of you non-halting is avoided.