In a concrete category (i.e., where the morphisms are functions between sets), I define a base of an object $A$ to be a set of elements $M$ of $A$ such that for any morphisms $F,G:A\to B$ that coincide on $M$, we have $F=G$.

Question: Is there an established name for a base in that sense?

Examples: In the category of vectors spaces, generating sets are bases. In the category of sets, $A$ is the only base of $A$.

Note: The above definition does not really need a concrete category (an initial object is enough), but I decided to formulate it in a concrete category for simplicity.

  • 2
    $\begingroup$ Surely for vector spaces, but also in general: what a bout "a spanning subset" or, as you suggested implicitly, "a generating set"? $\endgroup$
    – Uri Bader
    May 15, 2021 at 11:09
  • 3
    $\begingroup$ Doesn't 'generating set' fit your requirements? I see no reason to invent anything fancier. $\endgroup$
    – Denis T
    May 15, 2021 at 11:41
  • 4
    $\begingroup$ You don't need to put the identity in as a generator for unital algebras. It's part of the signature. $\endgroup$ May 15, 2021 at 12:34
  • 1
    $\begingroup$ Another example why generating set does not fit: In the category of topological space with continuous functions, a "base" would be a dense subset. I don't think one would usually say that the dense subset "generates" the space. $\endgroup$ May 15, 2021 at 12:34
  • 1
    $\begingroup$ If you allow nonunital homomorphisms then 1 is not part of the generating set because you are just working in the category of not necessarily unital rings and so you would have to include it as a generator. But then I'd don't think 0 would be a base because you can map $\mathbb C$ to itself by the zero map and the identity and these both agree on zero and are X- algebra homomorphisms in the nonunital sense in that f(cx)=cf(x) for all complex numbers c. But you have to define your morphisms clearly $\endgroup$ May 15, 2021 at 13:52

2 Answers 2


The term "base" should not be used, since, as you say, you are actually generalizing the notion of a generating set.

It is an epi-sink, also known as jointly epimorphic family. See Joy of Cats, Definition 10.62 and (dual of) Definition 10.5. A family of morphisms $(f_i : A_i \to A)$ is called an epi-sink when for $u,v : A \to B$ we have $\forall i (u \circ f_i = v \circ f_i) \implies u=v$. When the coproduct $\coprod_i A_i$ exists, this means that we have an epimorphism $\coprod_{i \in I} A_i \to A$.

If you have a terminal object $1$, morphisms $1 \to A$ are called global elements, and we can look at epi-sinks consisting of global elements of $A$.

For many categories, though, global elements are not enough. When we have a forgetful functor $U$ to $\mathbf{Set}$ with a left adjoint $F$, we have $U(-) \cong \hom(F(1),-)$, so that elements of the underlying set can be seen as morphisms on $F(1)$, and we can talk about epi-sinks on $F(1)$.

But the most general form does not put any restrictrions on the domains at all.

  • 10
    $\begingroup$ +1, but “jointly epimorphic family” is a much more common term for this than “epi-sink”, at least in my experience. $\endgroup$ May 15, 2021 at 22:01
  • 1
    $\begingroup$ This is true (I have added it), perhaps also because "jointly monomorphic family" is much more common than "mono-source" (as defined in Joy of Cats), since "source" already has a different meaning in category theory. $\endgroup$ May 15, 2021 at 23:12
  • $\begingroup$ There is no Definition 10.26 in the linked Joy of Cats. Maybe it's a typo in the number? $\endgroup$ May 16, 2021 at 7:52
  • $\begingroup$ Yes I corrected the typo. $\endgroup$ May 16, 2021 at 9:21

At least in the context of von Neumann algebras, separating is used for this concept. Confer [Takesaki], Definition II.3.16 (slightly reformulated):

Definition. Let $\mathcal M$ be a von Neumann algebra on $\mathfrak H$. A subset $\mathfrak U$ of $\mathfrak H$ is called separating for $\mathcal M$ iff for all $a\in\mathcal M$, $a\xi=0$ for all $\xi\in\mathfrak U$ implies $a=0$.

(But note also the definition of a separating set in nLab which is related by a different concept.)

[Takesaki] Takesaki, Masamichi, Theory of operator algebras I, New York, Heidelberg, Berlin: Springer-Verlag. VII, 415 p. DM 79.00; $ 44.30 (1979). ZBL0436.46043.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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