In Munkres's Topology textbook, it says,


A subbasis S for a topology on X is a collection of subsets of X whose union equals X.

The topology generated by the subbasis S is defined to be the collection T of all unions of finite intersections of elements of S.

I am having a very hard time understanding the first sentence. Let's say $X = \{1, 2, 3\}$. Then let $A = \{\{1\}, \{2\}, \{3\}\}$, or $A = \{\{1, 2\}, \{3\}\}$. Either way, $A$ is a collection of subsets of $X$, and the union of the elements in $A$ is $X$, but $A$ does not seem like a subbasis here...

What did I misunderstand here? What should be the correct interpretation of "a collection of subsets of X whose union equals X"?

Thank you in advance!


So turned out $A$ is a subbasis in both cases, thanks @Surb for your comment.

But here is what I don't understand (which made me mistakenly think A may not even be subbasis in the first place). If $A = \{\{1\}, \{2\}, \{3\}\}$, then wouldn't "the collection T of all unions of finite intersections of elements of $A$" be {$\varnothing$}, since any intersection of any two elements in $A$ is empty? But since T is a topology, T must also contain $X$, but, in this case, it does not?

  • 4
    $\begingroup$ Both work and will provide different topologies on $X$. $\endgroup$
    – Surb
    Mar 9, 2021 at 20:34
  • 3
    $\begingroup$ Why do you say $A$ does not seem like a subbasis here? $\endgroup$ Mar 9, 2021 at 20:38
  • $\begingroup$ @Tanner I added my thoughts in the post. Thanks! $\endgroup$
    – Erin
    Mar 9, 2021 at 21:02
  • 1
    $\begingroup$ Note in response to edit: the intersection of two elements of $A$ is not necessarily empty, because for example $\{1\}\cap\{1\}=\{1\}$ $\endgroup$ Mar 9, 2021 at 21:04
  • $\begingroup$ Ah, that clarifies everything. Thank you! @Tanner $\endgroup$
    – Erin
    Mar 9, 2021 at 21:08

3 Answers 3


The first "subbase" is {{1}, {2}, {3}}. The set of all possible unions are {{1}, {2}, {3}, {1,2}, {1, 3}, {2, 3}, {1, 2, 3}}. The set of all possible intersections is {{1}, {2}, {3}, {}}.

The set of all possible unions and intersections is {{}, {1}, {2}, {3}, {1,2}, {1, 3}, {2, 3}, {1, 2, 3}}. That is a topology, in fact it contains all subsets, the "discrete topology" for {1, 2, 3}.

The second subbase is {{1, 2}, {3}}.

The set of all possible unions is {{1, 2}, {3}, {1, 2, 3}}.

The set of all possible intersections is {{1, 2}, {3}, {}}.

The set of all unions and intersections is {{}, {3}, {1, 2}, {1, 2, 3}}. Again, that is a topology for {1, 2, 3}.

If you could not see that, are you clear what a "topology" for a set is? A topology for a set, X, is a collection of subsets or X that has four properties:

  1. It contains X itself.
  2. It contains the empty set, {}.
  3. It contains all unions of its sets.
  4. It contains all finite intersections of its sets.

What you’re missing is that a finite subcollection of $A$ can be a collection with only one element; in your example, for instance, $\big\{\{1\}\big\}$ is a finite subcollection of $A$.

If $\mathscr{C}$ is any family of subsets of a set $X$, by definition

$$\bigcap\mathscr{C}=\{x\in X:\exists C\in\mathscr{C}(x\in C)\}\,;$$

$\bigcap\mathscr{C}$ is the set of all things that belong to at least one member of $\mathscr{C}$. In your example $\big\{\{1\}\big\}$ is a finite subset of $A$, and and its intersection is the set of all members of $\{1,2,3\}$ that belong to some member of the collection $\big\{\{1\}\big\}$. The only member of that collection is $\{1\}$, and the only thing that belongs to it is $1$, so $\bigcap\big\{\{1\}\big\}=\{1\}$.

More generally, if $S$ is any set, $\bigcap\{S\}=S$: the only member of the collection $\{S\}$ is $S$, and the set of its members is just $S$ itself.

  • $\begingroup$ Hi professor Scott, could I ask your assistance here, please? $\endgroup$ Mar 9, 2021 at 21:22

Munkres has a slightly modified definition of subbase, from my perspective.

For any family $\mathcal{S}$ of subsets of $X$ there is a unique smallest topology $\mathcal{T}(\mathcal{S})$ that contains $\mathcal{S}$, which we can define abstractly as the intersection of all topologies on $X$ that contain $\mathcal{S}$; we at least have one such topology (the discrete one, which is just all subsets of $X$) and any intersection of topologies is again a topology.

But there is also a less abstract way to define $\mathcal{T}(\mathcal{S})$ which is quite natural once, you have seen bases of topologies:

Let $\mathcal{S}'$ be the set of all intersections of finite subfamilies of $\mathcal{S}$. (if $\mathcal{N} \subseteq \mathcal{S} \subseteq \mathscr{P}(X)$, we define $$\bigcap \mathcal{S} = \{x \in X\mid\forall O \in \mathcal{N}: x \in O\}\tag{1}$$

as usual in set theory and only take those intersections for finite subfamilies $\mathcal{N}$. Now topologies are always closed under finite intersections so if $\mathcal{T}$ is any topology that contains $\mathcal{S}$, it will also contain $\mathcal{S}'$, defined by all finite intersections from $\mathcal{S}$ in the above sense.

By the notion of "void truth" it holds that $\bigcap \emptyset = X$ as we have a universal quantifier over the $\mathcal{N} = \emptyset$ in $(1)$. So we get "for free" that $X \in \mathcal{S}$, as the empty subfamily is surely finite. Also $\mathcal{S} \subseteq \mathcal{S}'$ as when $S \in \mathcal{S}$, $\{S\}$ is also a finite subfamily of $\mathcal{S}$ and $\bigcap \{S\} = S \in \mathcal{S}'$ by the definition. But of course also $S_1, S_2 \in \mathcal{S}$ implies $S_1 \cap S_2= \bigcap \{S_1,S_2\} \in \mathcal{S}'$ and likewise for intersections of $3$,$4$ etc. members. All of these must be in any topology containing $\mathcal{S}$, so in $\mathcal{T}(\mathcal{S})$ in particular.

Anyway, we then recall the criteria that Munkres gives for a family $\mathcal{B}$ to be a base for some topology:

  1. $\bigcup \mathcal{B} = X$
  2. $\forall B_1, B_2 \in \mathcal{B}: \forall x \in B_1 \cap B_2 \exists B_3 \in \mathcal{B}: x \in B_3 \subseteq B_1 \cap B_2$

and if these conditions are satisfied, the set of all unions of subfamilies $\mathcal{B}' \subseteq \mathcal{B}$, where $$\bigcup \mathcal{B}' = \{x \in X\mid \exists B \in \mathcal{B}: x \in B\}\tag{2}$$ forms the unique minimal topology that has $\mathcal{B}$ as a base.

Now we easily check that our $\mathcal{S}'$ defined from $\mathcal{S}$ indeed satisfies 1. and 2. That 1. holds follows from $X \in \mathcal{S}'$ trivially. That 2. holds is because $\mathcal{S}'$ is closed under finite intersections: given $B_1 = \bigcap \mathcal{N}_1$ and $B_2 = \mathcal{N}_2$, with $\mathcal{N}_1, \mathcal{N}_2 \subseteq \mathcal{S}$ finite, we just pick for any $x$ in their intersection $B_3 = \bigcap (\mathcal{N}_1 \cup \mathcal{N}_2) \in \mathcal{S}'$ (as a union of two finite sets is finite).

So $\mathcal{S}'$ forms a base for $\mathcal{T}(\mathcal{S})$ by minimality and the topology is just the set of all unions of (finite intersections from $\mathcal{S}$).

This $\mathcal{S}$ is traditonally called the subbase for this minimal topology $\mathcal{T}(\mathcal{S})$ generated by $\mathcal{S}$, so just another word for generating set: we just specify (as minimally as possible, usually) a collection of sets that has to be open to define a topology. This happens quite a lot in topology. The order topology is one example : we just specify that all sets of the form $\{x\mid x < a\}$ and $\{x\mid x > a\}$ where $a \in X$ have to be open, so these form a subbase for the order topology. (But not yet a base).

Munkres almost does the same thing but seems not to want to rely on the void truth convention to ensure that $X \in \mathcal{S}'$, but just demands as a condition that $\bigcup \mathcal{S} = X$ so that 1. is also satisfied for the base $\mathcal{S}' \supseteq \mathcal{S}$. The rest of the arguments then stay the same.

In practice this doesn't make much difference as standard subbases automatically satisfy that condition (the order topology subbase on $X$ only does if $|X| \ge 2$) and the set $\{p_i^{-1}[O]\mid i \in I, O \subseteq X_i \text{ open }\}$ for the product topology on $\prod_{i \in I} X_i$ etc.

The main reason why subbases got their separate name is that the Alaxander subbase lemma (or theorem) can be used to prove compactness of a space: $X$ is compact iff every open cover with elements from $\mathcal{S}$ has a finite subcover. This allows for efficient proofs of Tychonoff's theorem and other applications.

SO for me a subbase can be any family, but Munkres wants to avoid logical discussions and demands the union be $X$. But the basic idea stays the same: specify a small set that generates the topology.


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.