40

The problem statement, all variables and given/known data:
Show that if $X$ is a subset of $M$ and $(M,d)$ is separable, then $(X,d)$ is separable. [This may be a little bit trickier than it looks - $E$ may be a countable dense subset of $M$ with $X\cap E = \varnothing$.] Definitions Per our book:

A metric space $(M,d)$ is separable if there exists a countable dense $E$ contained in $M$.

$E$ contained in $M$ is dense if $\forall m\in M$, $\forall ε>0\in\mathbb R$, $\exists e\in E$ s.t. $d(m,e) < ε$

The attempt at a solution

My best attempt was doomed from the start, because I don't quite understand the hint. My thought process went as follows:

Since $X$ is a subset of $M$, $\forall x\in X, x\in M$. Thus, since $E$ is dense in $M$,

$\forall x\in X$, and ε > 0, $\exists e\in E$ st $d(x,e)<ε $.

At this point, I was done, because the set of $e$'s satisfying the above, is a subset of $E$, a countable set. So a subset of a countable set is dense in $X$, and $X$ is separable. This is incorrect, but I cannot see why. Any help clearing up the confusion would be greatly appreciated. Thanks!

Edit: I wish I could upvote all of you for your help! I really appreciate the speedy replies and attempts to make this information clear to me.

Patrick
  • 675
  • 2
    The problem with your argument is that these points $e$ you find may not be in $X$. To say that $X$ is separable means it CONTAINS a subset which is countable and dense in $X$. \ You will have to find a new countable subset of $X$ which is dense in $X$, by modifying $E$ somehow. (I hope this is a useful hint.) – GCD Oct 06 '13 at 18:37
  • I don't think I follow. How could X these points not be in X, if they are in E, which is a subset of X? – Patrick Oct 06 '13 at 18:39
  • 1
    $E$ is a subset of the big space $M$, not necessarily of $X$. $X$ is also a subset of the big space $M$, but $X$ and $E$ may have no points in common. – GCD Oct 06 '13 at 18:39
  • Answered in this, but can be proved easily from definitions. – leo Oct 06 '13 at 18:41
  • 1
    $E$ and $X$ are both subsets of the same large space $M$. They need not have any points in common. For example, $M$ could be the real line, $E$ could be the rational numbers (dense in $M$), and $X$ could be the irrational numbers. $X$ is indeed separable, but in this case a countable dense subset of $X$ can't be made up of points in $E$, because none of those points are in $X$. – GCD Oct 06 '13 at 18:43
  • @leo, I appreciate your linking me to another post, but most of the terms in that post, my class has not covered. This is a math for economists course taught to first year graduate students. Our professor has picked certain aspects of real analysis to cover. Subspaces, countable bases, and most everything used in the proof has not been used in my class. Again, I know this can be proved easily. My question is that I do not see why my answer is wrong. – Patrick Oct 06 '13 at 18:46
  • @GCD Yes, I know that E and X need not have any elements in common. However, it is entirely impossible that there exists an element in X and epsilon > 0 such that there is no e in E satisfying d(e,x)<epsilon. So, what subtlety of the definition am I missing that prevents me from claiming a subset of E is dense in X? I guess I'm confused as to why I need to prove this subset exists, when density of E in M seems to guarantee it? – Patrick Oct 06 '13 at 18:48
  • Ok, I'll add an answer following the definitions you gave. In a few minutes. – leo Oct 06 '13 at 18:48
  • @user99181 Brian Scott has written a nice answer, but I'll try to address your question. To say "$A$ is dense in $B$" presumes that $A$ is contained in $B$ (see your own definition above). So when you say "a subset of $E$ is dense in $X$", you are presuming at least that E and X contain some points in common, which is not necessarily true. For example, no subset of the rationals is dense in the irrationals, because there are no rationals in the irrationals. – GCD Oct 06 '13 at 18:52
  • @GCD I saw Brian's answer, and I'm trying to get through it and understand it. It does not help that the course I'm in just picks and chooses bits of real analysis to teach, and forgets the rest. We say R is dense in Q. So for ANY two real numbers, there is a Q between them. Thus, if I have two irrationals, there is a Q between them, and I'm done. Clearly the irrationals do not contain the rationals, but they are still dense in a subset of them. What am I missing? – Patrick Oct 06 '13 at 18:57
  • @user99181 Look at your own definition of "dense" in a metric space; the word "between" has nothing to do with it in general metric spaces. If $A$ is a subset of $B$, then $A$ is dense in $B$ if for any point in $B$, we can find points in $A$ arbitrarily close to it. It's true that any irrational has rationals arbitrarily close to it, but those rationals are not within the set of irrationals. Saying $A$ is dense in $B$ presumes that $A$ is inside $B$, as you yourself wrote in the definition in your original post. – GCD Oct 06 '13 at 18:59
  • @gcd I was told that "between" and the distance definition were equivalent. I guess that is not true? Regardless, since R is dense in Q, for ALL real numbers, including irrationals, I can find a q in Q st d(q,r)<epsilon. So, how does limiting R to its subset of irrationals change this? – Patrick Oct 06 '13 at 19:02
  • Also, you seem confused about notation: the thing to say is that "$\mathbb{Q}$ is dense in $\mathbb{R}$", not vice versa. If it helps, I'll try to write yet another answer without using "open sets" at all, though Brian Scott's and kahen's are both fine. – GCD Oct 06 '13 at 19:04
  • @GCD my bad on the notation. I don't want you to go to the trouble of writing up another answer. I guess I just don't understand why, when Q is dense in R, if I have r in R st r is irrational, and any epsilon > 0, I can find a q in Q st d(q,r) < epsilon, but when I restrict R to just the irrationals, I no longer can do this? Wouldn't I just pick the same q as before? – Patrick Oct 06 '13 at 19:08
  • The idea is we would like to consider $X$ as its own metric space, without reference to the ambient larger space $M$. To say that $X$ is separable is solely a property of $X$, whether it lives in $M$ or not. So to show that $X$ is separable we need to find a countable dense SUBSET of $X$. – GCD Oct 06 '13 at 19:13
  • If I can't use the fact that X lives in M, then what is the purpose of knowing M is separable? I've been working on this problem for 4 hours today, and have now read 3 different solutions. None of them make sense to me. I'm ready to just throw in the towel. – Patrick Oct 06 '13 at 19:15
  • Take a look at the solution I wrote, maybe it gives another perspective. You use the subset $E$ to build a different countable dense subset of $X$ that lives inside $X$. These issues are always tricky the first time! – GCD Oct 06 '13 at 19:18

4 Answers4

57

Note that it’s very important here that you’re working in a metric space, because the statement isn’t true in topological spaces in general.

Since $X$ is a subset of $M$, $\forall x\in X, x\in M$. Thus, since $M$ is dense in $E$,

$\forall x\in X$, and $\epsilon > 0$, $\exists e\in E$ st $d(x,e)< \epsilon $.

‘Since $M$ is dense in $E$’ doesn’t make sense. First, you never defined $E$. I can guess that it’s supposed to be a countable dense subset of $M$, but then your statement is just backwards: $E$ is dense in $M$. In any case, finding points of $E$ near $x$ doesn’t help to show that $X$ is separable: to show that you must find a countable subset of $X$ that is dense in $X$, and $E$ might be completely disjoint from $X$. For a concrete example of this possibility, let $M=\Bbb R$ with the usual metric, and let $X$ be the set of irrational numbers. We know that $\Bbb R$ is separable, because $\Bbb Q$ is a countable dense subset of $\Bbb R$. The theorem that you’re to prove says that $X$ is also separable, i.e., that there is some countable set $D$ of irrational numbers that is dense in $X$, but $D$ certainly can’t be $\Bbb Q$, our familiar countable dense set of reals: no member of $\Bbb Q$ is even in $X$.

One way to prove the theorem is to show that if $E$ is a countable dense subset of $M$, then $$\mathscr{B}=\{B(e,r):e\in E\text{ and }0<r\in\Bbb Q\}$$ is a base for $M$, meaning that if $x\in M$, and $U$ is an open set containing $x$, then there is some $B(e,r)\in\mathscr{B}$ such that $x\in B(e,r)\subseteq U$. Note that $\mathscr{B}$ is a countable family of open balls. Then let $$\mathscr{B}_0=\{B\cap X:B\in\mathscr{B}\text{ and }B\cap X\ne\varnothing\}\;,$$ and show that $\mathscr{B}_0$ is a base for $X$. Since $\mathscr{B}$ is countable, so is $\mathscr{B}_0$. Finally, for each $B\in\mathscr{B}_0$ pick one point $x_B\in B$, and let $D=\{x_B:B\in\mathscr{B}_0\}$; $D$ is countable, and it’s not too hard to show that it’s dense in $X$ and hence that $X$ is separable.

Brian M. Scott
  • 616,228
  • I follow along until you introduce B0. That theorem is listed immediately before this question in my book. That if (M,d) is a separable metric space, then there exists a countable collection of open sets such that every open subset of M is a union of some open sets from the collection. I don't understand why any of the B's can be guaranteed to have a non empty intersection with X? – Patrick Oct 06 '13 at 19:12
  • 1
    @user99181: Because $E$ is dense in $M$. Let $x\in X$ and $n\in\Bbb Z^+$; there is some $e\in B\left(x,\frac1n\right)\cap E$. Then $x\in B\left(e,\frac1n\right)\cap X$, and $\frac1n$ is certainly rational, so $B\left(e,\frac1n\right)\in\mathscr{B}_0$. – Brian M. Scott Oct 06 '13 at 19:19
19

EDIT: Fixed a mistake.

Here is a slightly different answer, though of course basically equivalent to the other two.

$X$ is a subset of $M$. $E$ is a countable dense subset of $M$. We would like to find a countable dense subset of $X$.

The problem, as hashed out in the comments above, is that we would like to use $E$ as our countable dense subset of $X$, but the points of $E$ may not actually belong to $X$.

So one thing we can do is this: Every point $e$ in $E$ has some distance $$ d(e, X) = \inf\{ d(e, x) : x\in X\} $$ i.e. how far it is from the set $X$.

For every $e$ in $E$, choose points $a_n$ in $X$ whose distance from $e$ is, say, less than $d(e,X) + 1/n$. (I may not be able to get $a$ exactly distance $d(e,X)$ from $e$, but I can get close.)

Now, let $A$ be the set of all $a_n$ that I made, countably many from each $e$ in $E$. (Okay, some $a_n$ may belong to multiple $e$'s, but so what.) $A$ is in $X$, by definition, and $A$ is countable, as $E$ was (countable union of countable sets is countable).

So I just have to show that $A$ is dense in $X$. Let $x$ be any point of $X$, and take any $\epsilon>0$. Then there is some $e\in E$ within $\epsilon/3$ of $x$, i.e. $d(x,e)<\epsilon/3$. That means $d(e, X)\leq \epsilon/3$, so there is some $a\in A$ such that $$ d(e,a) < d(e,X) + \epsilon/3 \leq 2\epsilon/3. $$

So $$ d(x,a) \leq d(x,e) + d(e,a) < \epsilon/3 + 2\epsilon/3 = \epsilon. $$

So $A$ is dense in $X$.

GCD
  • 906
  • For every e in E, choose a point a in X whose distance from e is, say, less than 2d(e,X). (I may not be able to get a exactly distance d(e,X) from e, but I can get close.) This is the part that threw me for a loop in the solutions I found earlier.

    The line above we said d(e,X)=inf{d(e,x): x in X}, then for every e, I can pick an a in X st d(e,a) is exactly d(e,X). Namely, I pick the point in X st d(e,a)=inf{(d(e,x): x in X}.

    – Patrick Oct 06 '13 at 19:20
  • (Sorry I just edited to fix a mistake; I'm an inveterate editor.) But nope, infimums are not always achieved. There might not be a point which actually realizes the minimum distance. For example, what is the distance from the point $0$ to the set $(0,\infty)$ in $\mathbb{R}$? The distance is zero, but there is no point in $(0,\infty)$ which actually has distance zero from $0$. – GCD Oct 06 '13 at 19:25
  • Okay, the part about the infimum and picking an a in X makes sense. Since I have this infimum of the distance, then, adding any amount > 0 to d(e,X), I guarantee the existence of an a in X st d(e,X)<= d(e,a) < d(e,X)+epsilon. Right? – Patrick Oct 06 '13 at 19:31
  • 1
    Hey @GCD, OP here, but from a different computer. Your proof makes the most sense to me...I can't follow the others. However, I'm hung up on the part where we show A is dense in X. You have d(e,a) < d(e,X) + ε/3, but I don't see why this is always true. All we have to go on is that d(e,a) < d(e,X)+1/n <= ε/3+1/n. I don't see how to guarantee 1/n <= ε/3? Suppose that the set of an's is a singleton (i.e. for all e in E, the a in X that minimized distance was the same. Then 1/n is just 1.) – Patrick Oct 06 '13 at 21:09
  • 1
    Given $e\in E$ I would like to just choose $a\in X$ associated to it so that $d(e,a) = d(e,X)$. Unfortunately this may not be possible. So I do the next best thing, which is to associated to $e$ a countable set of $a_n$'s getting arbitrarily close to the minimal distance. In the part of the proof you ask about, I simply choose $n$ large so that $1/n<\epsilon/3$; in other words so that $a_n$ comes within $\epsilon/3$ of minimizing the distance. – GCD Oct 08 '13 at 03:46
  • @PatrickKriebel If you liked this answer or any of the other answers provided by the other users then you can click the check mark beside the answer to "accept" it. If you have 15 rep or above, then you can even upvote an answer to show gratitude; you can upvote multiple answers. This will help the user and yourself – leo Nov 23 '13 at 15:47
9

Hint (for a different proof strategy): Show that $(M,d)$ is second countable, i.e. there is a countable collection $\mathcal B = \{U_n : n \in \omega\}$ of open sets of $M$ such that any open set $U$ in $M$ can be written as a union of elements of $\mathcal B$. Then show that that $X$ also is second countable. And finally show that second countable implies separable.

kahen
  • 15,760
  • 3
    I should add that this relies heavily on $M$ being a metric space. In the more general setting of point set topology, separability does not imply second countability. Not even when the space is first countable as well. – kahen Oct 06 '13 at 18:54
  • I have, from a theorem in our book, that M is second countable (it doesn't use that word, but the theorem states exactly the same definition you give for second countable). Even knowing that, how can I show X is second countable? How do I know X isn't empty, or even contains an open set? – Patrick Oct 06 '13 at 19:17
  • @user99181 That doesn't matter. A $U \subseteq X$ is $(X,d)$-open if and only if there is an $(M,d)$-open $V \subseteq M$ such that $U = X \cap V$. And if $X$ is empty, then it's trivially separable. – kahen Oct 06 '13 at 19:22
  • Hello. I followed your answer, but got stuck at showing that X is also second countable. Can you hint me towards an answer? –  Mar 04 '18 at 21:39
4

This answered here which is the approach in the Brian M. Scott's answer.

Since you pointed out that you are not used to the terms used, and that this came from a course of math for economists, I'll try to give an answer using the definitions that you provided.

First, as an advice, when you are asked to proof something, at least at the beginning, it is better to look at the definitions and write down what you have to prove for the object you are supposed to prove things about.

In this case you have to prove that $(X,d)$ is separable.

What is $(X,d)$? It is the metric space formed by the subset $X$ of $M$, in which we measure distances just as we do in $M$.

What you have to prove about $(X,d)$? You have to prove that there exist a countable subset $E$ of $X$ such that $E$ is dense in $X$.

But you know that $(M,d)$ is separable, so there exist a countable subset $D$ of $M$, say $$D = \{x_1,x_2,x_3,\ldots\},$$ which is dense in $M$.

Now, let's enumerate the positive rationals as $$\Bbb Q \cap (0,\infty) = \{r_1,r_2,r_3,\ldots\}.$$

Define $$\Delta = \{(i,j) : B(x_i,r_j)\cap X\neq\emptyset\}.$$ $\Delta$ is not empty because otherwise $D$ is not dense in $M$.

Then, for each $(i,j)\in\Delta$ there is an $e_{(i,j)}\in B(x_i,r_j)\cap X$. Define then $$E = \{e_{(i,j)}: (i,j)\in \Delta\}.$$

This $E$ is a countable dense subset of $X$, so we are done.

If this last statement is not clear, let me know it and I'll elaborate on it.

Edit Indeed, Let $x\in X$ and $\epsilon\gt 0$. There's an $r_j$ such that $r_j\lt \epsilon/2$. Since $D$ is dense in $M$ there exist some $x_i\in D$ such that $d(x,x_i)\lt r_j$. So both $x$ and $e_{(i,j)}$ are elements of the ball $B(x_i,r_j)$, therefore $$d(x,e_{(i,j)})\leq d(x,x_i) + d(x_i,e_{(i,j)}) = 2r_j \lt \epsilon.$$

leo
  • 10,433