We have seen above that
any finite set of real numbers,
the set ,
the set of all real numbers,
can be viewed as the set of limitpoints for some sequences. However, the set cannot be the set of limitpoints of any ANY sequence.
This begs for the question: If we come across with a set of real numbers, can we decide whether there exists some sequence such that . The following proposition gives an important condition for being limitset.
[How limitsets look like] Let be a sequence. Suppose that is a convergent set of limitpoints of tending to the number . Then is also a limitpoint of .
Proof: We already had versions of this proposition: in the example when we showed that is not a limitset and in the proof of the fact that Limsup is always a limitpoint. We use exactly the same idea. Since are limitpoints, we have a sequence such that So, by the Sum Rule . That is it.
By the previous proposition any limitset must have the following property: Now matter how we choose a convergent sequence from the limit of is in the set . This was of course true for the three examples in Kind Reminder 3.3.1. Also, it does not hold for the set of rational numbers and the set of irrational numbers as well (yes, you can approximate irrational numbers by rationals AND any rational number by the sequence of irrationals : The following definition is extremely important.
A set is closed if for any convergent sequence , provided that for any , is in .
The intervals are closed sets. Finite sets are closed sets. The empty set is a closed set by convention (and yes, the empty set is the set of limitpoints of the sequence ). The set of all real numbers, and the set of all positive real numbers are closed as well. The set is also closed.
Let be the set of all real numbers in between and such that all the digits of
are either or . This is a closed set that is
” not countable” and does not contain intervals.
The following two propositions help you to find even more complicated closed sets.
Let be a sequence of closed sets. Then their intersection is a closed set as well.
Proof: We only need to show that if is a convergent sequence tending to such that for any , then as well. This is almost trivial. For any and , so since is closed. The fact that is an element of all the ’s means exactly that .
If and are both closed sets then is a closed set as well.
Proof: Let be a convergent sequence tending to such that . We need to show that . Clearly, either there is a subsequence so that all the ’s are in or there is a subsequence so that all the ’s are in . The limit of this subsequences is of course. So is either in or in , that is .
If is an arbitrary sequence in , then is a closed set.
Proof: This seems like a new proposition. It is not. It is just Proposition 3.3.1 in a new cloth. We proved that the limit of a convergence sequence of limitpoints of is again a limitpoint. This means that the set of limitpoints of is a closed set, not less, not more.
Let be a closed non-empty set inside the interval . Then there exists a sequence so that the The theorem actually holds for all closed sets but the proof is a bit more (not much, a bit, it will be a workshop question) complicated.
Proof: The proof is given by an algorithm. We take first the ten -intervals, then the hundred -intervals and so on. So, we will have the system of intervals . Our algorithm works the following way. Let us pick the first interval that contains an element of and let this element be . Then go further, pick the next interval that contains an element of and let this element be . Continuing the process we will obtain a sequence . Then . Why? We need to show two things.
All elements of are limitpoints of .
If then is not a limitpoint of .
The second statement is trivial. Indeed, if a convergent subsequence tending to then all the are in our closed set , so by the very definition of closedness must be in as well. The first statement is a bit more trickier, but not much. Let . Then is in the -interval , in the -interval , in the and so on. When our algorithm run we cannot avoid this intervals. We must pick a point from each of them. So we will have a subsequence such that . That is , hence converges to .
We have seen many kind of closed sets. Simple ones, complicated ones, finite ones and infinite ones. There is a very simple way to construct closed sets from arbitrary sets. As we have seen above for any closed set , there exists a sequence of elements of such that the set of limitpoints of the sequence is exactly . So, it we start with this sequence, we can reconstruct . The following theorem is a variation of this theme.
Let be an arbitrary set. Let be the set of all points that can be attained as a limit of a convergent sequence . Then
The set contains .
The set is closed.
Any closed set that contains will contain as well. Hence is the smallest closed set containing and it is called the closure of .
Proof: Let . Then is a perfectly legitimate convergence sequence of elements of and its limitpoint is of course. So, . Now, we prove that is closed. Let be a sequence from tending to . We need to prove that . Since, is in , there exists such that . By the Sum Rule, the sequence is convergent and its limit is . Hence, is element of . Finally, suppose that , where is some closed set. We need to prove that if then . What does it mean that ? It means, that there exists a convergent sequence of elements of tending to . This convergent sequence is in , is closed, hence must be in .
Let be a closed set and . Suppose that . Then we say that is dense in .
is dense in .
The set of irrational numbers is also dense in .
The numbers in the form , where is an integer is dense in .
Every closed set contains a countable dense set.
Proof: As we have seen before, for all real number , there is a sequence of rationals tending to . Indeed, if then the sequence tends to . This observation takes care of (1) AND (3) as well!!! If is a rational number and is an irrational number, then is an irrational number. Hence if is a convergent sequence of rational numbers tending to , then is a convergent sequence of irrational numbers tending to . The statement (4) is just the reformulation of Theorem 3.3.1.