[bjornsing]

> The Turing machine has a tape of unbounded size so can’t be built simpliciter.

On the contrary, I think this is one of the advantages of Turing’s model: I can imagine standing there in my garage looking on as my universal Turing machine is running low on tape on the left side, and then simply attaching a new roll of fresh empty tape at the end, holding it as it is fed into the machine. :)

It’s simply the least leaky abstraction of computation.

[codeflo]

Exactly this. Unbounded doesn't mean infinite, and people are sometimes confused by the distinction.

[hardlianotion]

Including me. What is the difference?

[codeflo]

Some parts of mathematics deal with infinite sequences, that is, actually infinite lists of numbers. It's usually assumed, and important for analysis, that these numbers are considered to be "all there" right from the beginning. You can do operations like: Compute the limit. Add up all of its elements. Determine whether two sequences are identical for all entries after the trillionth.

I think this is often part of the misunderstanding when you stumble into a post by someone who's confused about 0.999... = 1. People sometimes write things like: "0.999... only moves closer and closer to 1, it never reaches 1." I think that highlights a deeper point than people usually give these comments credit for. The thing is, 0.999... doesn't "move" anywhere, it's considered a completed object right from the beginning.

Anyway, the point is that Turing machines are not like this at all. They only look at a fixed-size part of the tape during each step, from this follows that they have only used a finite amount of tape at each point of their execution.

So for any given (halting) computation, you don't actually need an infinite tape, you just need "enough", without changing the result. This is important because it makes Turing machines a model for practical computers. For example, the device you're reading this on has gigabytes of tape, and that's big enough for many, many, many kinds of computation.

[Dylan16807]

In a theoretical sense, an unbounded number is always finite.

In a practical sense, turing machines don't voraciously consume tape. Adding extra feet of tape gives you an exponential increase in what you can compute. So if you set up a program to be reasonably judicious with its tape use, you can just say that if it reaches an end you pause it for a day, head to the shop, and buy another reel. Big computations take a lot of time anyway.

[hardlianotion]

Any given number is always bounded. I am not sure it makes sense to talk about an unbounded number

[Dylan16807]

The size number of the unbounded tape size is always finite, is that better?

[hansvm]

The usual difference is just predicate ordering -- (1) for every program there exists a tape big enough vs (2) there exists a tape big enough for every program. In the first case, each individual (valid) program can get by with a tape of _some_ fixed length, but there's no bound on how big that requisite length might be. In the second case, since the tape requirements can be arbitrarily high you would need a legitimately infinite tape to service all possible programs.

IMO the given example muddies the waters a bit by conflating the conceptual tape a given machine is running on (which might be infinite for non-halting programs) with the physical tape you've used so far (for which it suffices for such tape to be finite at any fixed moment in time, though the amount needed might grow unboundedly as time increases).

[codeflo]

Note that tape usage typically depends on the input, so I would distinguish programs and computations (program + input).