While reading "Introduction to Topology and Modern Analysis" by G.F Simmons, in the chapter on Topological Spaces, I came across the following statement that seemed unusual. I feel it may be a vacuous truth but I don't know why. The statement is the following ad verbatim:

Let $X$ be a non-empty set. Let $\mathcal{S}$ be an arbitrary class of subsets of $X$.

If $\mathcal{S}$ is empty, then the class of all finite intersections of it's sets is the single element class $\{X\}$ and the class of all unions of sets in this class is $\{\varnothing,X\}$.

I do not follow this argument. So I would appreciate it if someone could throw some light onto this issue. If it is a vacuous truth, I would like to see how.

If this question has been asked before, I'd like to know under what title it lies in as I did not find it.

  • 2
    $\begingroup$ It seems that those are definitions, the same way as $\sum _{x \in S}$ is defined as $0$ if $S = \emptyset$. $\endgroup$ – Karolis Juodelė Feb 21 '13 at 8:48
  • $\begingroup$ As @KarolisJuodelė suggests, it is indeed a definition. See chapter 1, section 2 (page 12 in my edition) for definitions. $\endgroup$ – Jas Ter Feb 21 '13 at 8:53
  • 2
    $\begingroup$ Note that "in this class" is slightly ambiguous -- it's meant to refer to the single-element class $\{X\}$, but one could also interpret it as referring to $\mathcal S$ (in which case the statement would be wrong). $\endgroup$ – joriki Feb 21 '13 at 9:41
  • $\begingroup$ @joriki: I gave the statement ad-verbatim i.e. exactly as said, in the book. In this case, it refers to the former of what you said. $\endgroup$ – Gautam Shenoy Feb 21 '13 at 9:47

Both are instance of something being vacuously true, though one has to think a bit to see it.

Since $\mathcal{S}\subseteq\wp(X)$, for any $\mathscr{U}\subseteq\mathcal{S}$ we have

$$\bigcup\mathscr{U}=\{x\in X:\exists U\in\mathscr{U}(x\in U)\}\tag{1}$$


$$\bigcap\mathscr{U}=\{x\in X:\forall U\in\mathscr{U}(x\in U)\}\;.\tag{2}$$

Let’s take $(2)$ first. How could one conclusively demonstrate that some $x\in X$ did not belong to $\bigcap\mathscr{U}$? One would have to show that there was some $U\in\mathscr{U}$ such that $x\notin U$. If $\mathscr{U}=\varnothing$, it’s impossible even to find a $U\in\mathscr{U}$, let alone one that does not contain $x$, so it’s impossible to demonstrate that $x\notin\bigcap\mathscr{U}$. Thus, $x\in\bigcap\mathscr{U}$ for all $x\in X$ if $\mathscr{U}=\varnothing$, and hence $\bigcap\mathscr{U}=X$ when $\mathscr{U}=\varnothing$. In particular, if $\mathcal{S}=\varnothing$, then $\mathscr{U}\subseteq\mathcal{S}$ means that $\mathscr{U}=\varnothing$ and hence that $\bigcap\mathscr{U}=X$.

We deal with the union the same way. Given a particular $x\in X$, how do you show that $x\in\bigcup\mathscr{U}$? You show that there is a $U\in\mathscr{U}$ such that $x\in U$. If $\mathscr{U}=\varnothing$, you can’t even find a $U\in\mathscr{U}$, let alone one that contains $x$, so it must be that $x\notin\bigcup\mathscr{U}$. That’s true for all $x\in X$, so if $\mathscr{U}=\varnothing$, then $\bigcup\mathscr{U}=\varnothing$.

Note that in $(1)$ and $(2)$ I was careful to limit possible members of $\bigcup\mathscr{U}$ and $\bigcap\mathscr{U}$ to $X$. This didn’t matter much for $(1)$, since the argument given to show that no $x\in X$ belongs to $\bigcup\varnothing$ works equally well for any $x$ whatsoever, and we’d arrive at the conclusion that $\bigcup\varnothing=\varnothing$ even without the restriction. It matters a lot for $(2)$, however: without the restriction we end up concluding that $\bigcap\varnothing$ contains everything $-$ that it’s the ‘set’ of all sets, which doesn’t exist. By specifying that $\mathcal{S}$ is a class of subsets of $X$, Simmons is implicitly limiting the candidates for membership in $\bigcap\mathscr{U}$ (where $\mathscr{U}\subseteq\mathcal{S}$) to elements of $X$.

| cite | improve this answer | |
  • $\begingroup$ Thank you very much for the solution. I can therefore conclude that the truth is indeed vacuous. $\endgroup$ – Gautam Shenoy Feb 21 '13 at 9:41
  • $\begingroup$ @You’re very welcome. $\endgroup$ – Brian M. Scott Feb 21 '13 at 9:44
  • $\begingroup$ @Gautam: No mistake: I was talking about unions of subclasses of $\mathcal{S}$, and if $\mathcal{S}=\varnothing$, then the only subset of $\mathcal{S}$ is $\varnothing$, and $\bigcup\varnothing=\varnothing$. There is no $\mathscr{U}\subseteq\varnothing$ such that $\bigcup\mathscr{U}=X$. Simmons’s sentence is slightly ambiguous: when he says in this class, I took him to mean the class $\mathcal{S}$, when in fact he meant the class $\{X\}$. Of course the subclasses of that are $\varnothing$, whose union (as I explained) is $\varnothing$, and $\{X\}$, whose union is $X$. $\endgroup$ – Brian M. Scott Feb 22 '13 at 11:40
  • $\begingroup$ I apologize for my error. Incidentally, this is the only time I see union smaller than intersection :D $\endgroup$ – Gautam Shenoy Feb 23 '13 at 14:06
  • $\begingroup$ @Gautam: No need to apologize; the sentence really is a bit ambiguous. (Yes, I like that quirk too!) $\endgroup$ – Brian M. Scott Feb 23 '13 at 14:08

This fact is relying on the tacit assumption that the computation of the intersections is performed inside the lattice $\mathcal P(X)$, the set of all subsets of $X$ where the lattice operations are intersection and union (the elements in $\mathcal (X)$ are then ordered by set inclusion).

Now, in $\mathcal P(X)$ there is a largest element, namely $X$. The intersection of no sets is equal to that largest element. To see, that think of the property of intersection, that if $U\subseteq V$ where $U,V$ each is a family of subsets of $X$ then the intersection of all sets in $V$ is contained in the intersection of all sets in $U$. That is, if you intersect more sets you get a smaller subset. So, by that rational, the intersection of no sets at all should be the largest possible thing, thus it is $X$. (More generally, in any lattice the meet of the empty set if the largest element in the lattice (e.g., the infimum in $\mathbb R$ of the empty set does not exist, but if you adjoin $\infty $ to $\mathbb R$ in the usual way then the infimum of the empty set in $\mathbb R \cup \{\infty \}$ is $\infty $).)

So, starting with $S$ and forming all intersections of its elements gives you $\{X\}$. Now, forming all unions of the class $\{X\}$ gives rise to $\{\emptyset, X\}$. It is clear why $X$ is there. $\emptyset $ is there too for the dual reason that $X$ is the intersection in $\mathcal P(X)$ of no sets. The union of no sets should be the smalest possible thing, namely the empty set. (Similarly, the supremum in $[0,\infty )$ of the empty set is $0$).

It should be noted that these can be taken as arbitrary definition, adopted since it agrees with what you want computations to give, but this is in fact strictly a consequence of the definitions of infimum and supremum in lattices. You just need to realize that you are computing in the lattice $\mathcal P(X)$.

| cite | improve this answer | |
  • 1
    $\begingroup$ I never considered a lattice view. Thanks for that. Your analogy of infimum was indeed luculent. If only I could award two correct answers. $\endgroup$ – Gautam Shenoy Feb 21 '13 at 9:43
  • 1
    $\begingroup$ you're very welcome :) $\endgroup$ – Ittay Weiss Feb 21 '13 at 9:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.