I don't understand - the period-2 bulb is known in area and position - and can be considered algorithmically verified. And all the true shape Msets laserblaster and I are computing right now via interval arithmetics and cell-mapping are verified as well as the seed intervals enter a cycle for the interior. Given enough time and computer power, in principle every small square can be verified whether it lies fully in the interior of a hyperbolic component or fully outside the Mset (not allowed to touch the boundary itself though).

Or is there a definition of computability that collides with what I naively assume computability means? Or maybe there is more "interior" than just the hyperbolic components, wherever that then lies.

What about a square that's both in and out?

It would make sense if Penrose meant points in the set, rather than interior points.

Yes.

But maybe the notion is "there's (not yet) an algorithm to judge a general point" - and that reminds me of the halting problem in computer science: No general agorithm exists, but there might very well be one that could prove e.g. that any C++ compiler with no more than 1000 lines of code halts on every input.

Yes, the conjecture is "there is no halting algorithm that you input a single c and it tells you in M-set or not". c is assumed "computable" meaning there is an algorithm that computes digit after digit of c. For if c itself is uncomputable (like Chaitin's number) it's trivial. The discussion in Penrose's book is in context of Goedel theorem and Turing halting problem. So the idea is there is no better way than to iterate and hope it either converges or diverges but if neither it may stay bounded forever or not, no way to know.

Of course there is a better way, using human creativity. As trivial example if c is in main cardioid there is a better way than to iterate forever and give up: just prove it's inside the cardioid which has a simple computable shape. For pretty much any non-escaping orbit you compute you can figure out in which cardioid or pseudo-circle blob it is and then prove it's inside by solving a polynomial equation which is computable. So the non-computable points must be on the boundary. Penrose can't give an example of a non computable point. At least that's my summary of the book chapter, I hope I got it all right; the chapter is written as "computability for dummies" so I hope I'm not worse than dumb.

The paper Claude pointed at  defines "computability" in a more productive way AFAICU and seems to show M-set is computable in the sense you can make more and more accurate pictures of it and you can actually give guarantees on accuracy. But you already know that and are doing it.

Would be nice to come up with an algorithm (like an infinite series) that calculates a c such that no-one can figure out it it's in M or not. Proving uncomputability is an open problem so that would be too much to ask. Something like the first trancendental numbers that were proven; they were defined with an infinite series constructed explicitly so you can prove trancendentality.

Maybe pauldelbrot can come up with such a thing, at least I've never been able to ask a math question here that he could not answer

Random tidbit of useless information: the definition of "computability" is older than data processing machines. It used to mean a somewhat abstract thing. Then came computer science and overloaded the term with a more practical definition. It is possible to find a whole range of scientific papers with definitions inbetween these two ends (depending on what kind of abstract or concrete machine a paper is working with).

The meaning will shift again if and when quantum computers become relevant.

Good point, I thought computable alywas means Turing computable which includes quantum computers. Some more random reading in the Penrose book: if x is a computable real x==1? is uncomputable (undecidable) in general. For if your algorithm has computed 0.99..99 with a trilion 9's the trillionandoneth digit could be 8. So in this sense the unit circle is also not computable. Also in a foornote he writes someone told him non-computability of M-set was actually proven but no reference given.

I guess we're digressing far from rounding errors in polynomial evaluation...

If one have a year to check one point it is practicaly uncomputable

Get a faster machine and do it in 1 hr. There is no such thing as "practically uncomputable".