[sci.math] space filling

holsz@cadence.com (Wlodzimierz Holsztynski) (04/07/90)

In article <13170@thorin.cs.unc.edu> banks@pooh.cs.unc.edu (Father Banks) writes:
>Everyone has seen examples of space-filling curves. Wirth's text
>gives example code for generating two kinds that lie in the plane.
>
>Are there simple algorithms to generate space-filling curves
>in n dimensions? Even for n=3, hilbert 3curve code becomes
>unhappily complicated.
>
>David Banks

----------------------------

Let  f',f",g" : I --> I  and  g': I --> I^(n-1)  be continuous maps
such that their diagonal products  f = f'/\ f" : I --> I^2  and
g = g'/\g" : I --> n  are onto (are Peano "space filling" curves).
I assume that we already know how to construct  f  and  g  for n=2.
Let  h' = f' o g"  and  h" = f" o g".  Then

		H = g' /\ h' /\ h" : I --> I^(n+1)

is a continuous map of the interval I onto the (n+1)-dim cube.
By induction, we construct maps for n=2,3,4,.... In practise,
when dealing with a finite approximation, it's better to vary the
special coordinate of the construction when passing to the next
dimension. Don't choose the last one (as I did above) each time.

BTW, all this is well known. There are continuous maps of I onto
the hilbert cube Q too (and one may play some algorithmic games
in this infinite dimensional case too, if one insists). I = [0;1]
can be mapped onto a Hausdorff space X if and only if  X  is
compact, connected and locally connected. I don't remember the
author of this beautiful classical theorem :-( but the related
one:

	a complete connected and locally connected metric space
	is arcwise connected

was proved independently by Hahn and Mazurkiewicz (I think).

Remark. My symbol  /\  above is a weak imitation of the upper
case delta.

Enjoy
	Wlodek