Re: Accumulation points in metric space



[This (hopefully) corrects the first part of the proof
in my previous post.]

Dave L. Renfro wrote (in part):

If (X,d) is not separable, then there exists an uncountable
subset having no accumulation point in X. In fact, I believe
even more is true: If (X,d) is not separable, then there
exists an uncountable subset that is uniformly isolated
(i.e. there exists epsilon > 0 such that the distance
between each pair of distinct points in the uncountable
subset is at least epsilon).

I happen to be at home now, where my notes are. Here's
the result and proof.

THEOREM: If (X,d) is a non-separable metric space, then
there exists epsilon > 0 and an uncountable
subset E of X such that, for each pair of distinct
points x & y in E, we have d(x,y) > epsilon.

PROOF: I first claim there exists a positive integer n
such that for all countable subsets B of X there
exists x in X such that for all b in B we have
d(x,b) > 1/n. If not, then for all positive
integers k there exists a countable set B_k such
that for all x in X there exists b in B_k such
that d(x,b) <= 1/k. However, this contradicts the
non-separability of X, since UNION(k=1 to oo) B_k
is a countable dense subset of X. I will now construct
an appropriate set E, of cardinality aleph_1, by
transfinite induction over the ordinals less than
omega_1.

Choose any x in X and denote this point by x_1.
Let alpha > 0 be a countable ordinal and assume that,
for each ordinal beta < alpha, we have chosen a point
x_beta such that the distance between each pair of
distinct points in the set {x_beta: beta < alpha} is
greater than 1/n. Since {x_beta: beta < alpha}
is a countable subset of X, and there exist points
in X not belonging to this countable set (it trivially
follows from the non-separability of X that X is
uncountable), it follows from what I proved earlier
that there exists x_alpha in X such that each of the
distances between x_alpha and the points belonging
to {x_beta: beta < alpha} is greater than 1/n.
The set E = {x_beta: alpha < omega_1} is a set of
cardinality aleph_1 such that any two of its distinct
points are a distance of more than epsilon = 1/n
from each other.

NOTE 1: There does not exist a positive lower bound on epsilon
that uniformly works for all non-separable metric
spaces. This is because there exist non-separable
metric space of arbitrarily small diameter.
[Pick an uncountable set and define the distance
between each pair of distinct points to be an
arbitrarily small specified positive value.]

NOTE 2: In the case of the sequence space elle^infinity,
we can find a concrete example of cardinality c
with epsilon = 1: The collection of all sequences
whose entries are either 0 or 1.

NOTE 3: This theorem was given in a topology class I had
under Peter J. Nyikos in Spring 1987. A few years
ago (2001 or 2002), I came across the result in
Sierpinski's "General Topology" [1956 edition
translated by C. Cecilia Krieger and reprinted
by Dover Publications in 2000], Section 52,
Lemma, pp. 116-117. The proof Nyikos presented
and the proof in Sierpinski's book are both the
same as the proof above.

Dave L. Renfro

.



Relevant Pages

  • Re: Accumulation points in metric space
    ... (i.e. there exists epsilon> 0 such that the distance ... subset is at least epsilon). ... a countable subset B_k of X such that the distance ... follows from the non-separability of X that X is ...
    (sci.math)
  • Re: Point between 2 points
    ... the distance from A to C plus the distance from B to C is ... (lowercase epsilon). ... floating-points arithmetics at all... ... post represented using some funky uncommon representation? ...
    (comp.lang.java.programmer)
  • Re: Point between 2 points
    ... > the distance from A to C plus the distance from B to C is ... floating point inaccuracies. ... is within a certain error range of the desired result, ... epsilon EPS in my code. ...
    (comp.lang.java.programmer)
  • Re: Is there a standard name for this sort of connectivity?
    ... point y in a metric space X is a finite sequence of points in X ... successive points is less that epsilon. ... Here are some interesting locally connected aggregates ... connectivity, but in the noncompact case it is a weaker notion. ...
    (sci.math)
  • Re: Is there a standard name for this sort of connectivity?
    ... Is there a standard name for this sort of connectivity? ... point y in a metric space X is a finite sequence of points in X ... successive points is less that epsilon. ... Let S be the union of all lines thru with irrational slope. ...
    (sci.math)