Sequential and Limit Point Compactness
In metric spaces, compactness is equivalent to sequential compactness (every sequence has a convergent subsequence). In general topological spaces, these notions diverge. Understanding their relationships clarifies the role of countability axioms.
Definitions
A topological space is sequentially compact if every sequence in has a convergent subsequence.
A topological space is limit point compact (also called Bolzano--Weierstrass or weakly countably compact) if every infinite subset of has a limit point in .
A topological space is countably compact if every countable open cover has a finite subcover.
Relationships
For any topological space:
Compact countably compact: Every countable open cover is an open cover, so it has a finite subcover.
Countably compact limit point compact: Suppose is an infinite subset with no limit point. Then is closed (its complement is open since no point outside is a limit point). For each , there exists an open with (since is not a limit point of ). Choose a countably infinite subset . Then is a countable open cover of with no finite subcover (each is only covered by ).
Sequentially compact countably compact: Suppose is sequentially compact and is a countable open cover with no finite subcover. For each , choose . By sequential compactness, has a subsequence converging to some . Since covers , for some . But for , so no subsequence of can converge to unless the subsequence eventually enters -- which it does since is a neighborhood of . But for and , so eventually , contradiction.
Equivalence in Metrizable Spaces
For a metrizable space , the following are equivalent:
- is compact.
- is countably compact.
- is limit point compact.
- is sequentially compact.
- is totally bounded and complete (as a metric space).
We sketch the key implications not already proved.
(3 4 in metric spaces): Let be a sequence. If the range is finite, some value repeats infinitely often, giving a constant (hence convergent) subsequence. If the range is infinite, it has a limit point by (3). For each , the ball contains infinitely many . Inductively choose a subsequence ; then .
(4 1 in metric spaces): This is the most involved step and uses total boundedness. First, sequential compactness implies total boundedness: if not, there is an and a sequence with for all , having no convergent subsequence. Second, use the Lebesgue covering lemma (which holds in sequentially compact metric spaces) to convert any open cover to a finite subcover.
Counterexamples
-
Limit point compact but not sequentially compact: The ordinal space (first uncountable ordinal with the order topology) is limit point compact but contains the sequence which converges, so it is actually sequentially compact too. A better example: the product (with the product topology) is compact (Tychonoff) but not sequentially compact.
-
Sequentially compact but not compact: The ordinal space is sequentially compact (every sequence of countable ordinals is bounded, hence has a convergent subsequence) but not compact (the cover has no finite subcover).
-
Countably compact but not sequentially compact: is compact hence countably compact, but it is not sequentially compact.
In a first-countable space, sequential compactness and countable compactness are equivalent. In a second-countable space, all four notions of compactness (compact, countably compact, sequentially compact, limit point compact) are equivalent. This is why the distinctions are invisible in and other "nice" spaces.