Given a (small) category $C$, we can talk about the isomorphism classes of objects in $C$, and I was curious about how to define a morphism between $[X]$ and $[Y]$ for each pair of classes.

Denote $S=\{f \in C \mid \text{isomorphism}\}$. An intuitive way is to define the equivalence relation $\sim$ on $$\{f:X \rightarrow Y \mid X \in [X] , Y \in [Y]\}$$ for each pair of classes $[X]$, $[Y]$, where $(f_1:X_1\rightarrow Y_1) \sim (f_2:X_2\rightarrow Y_2) $ iff $\exists x_{12}:X_1\rightarrow X_2,y_{12}:Y_1\rightarrow Y_2 \;$in $S$ s.t. $f_2 = y_{12} \circ f_1 \circ x_{12}^{-1}$, i.e., we have the commutative diagram $\require{AMScd}$ \begin{CD} X_1 @>f_1>> Y_1\\ @A x_{12}^{-1} AA @VV y_{12} V\\ X_2 @>>f_2> Y_2. \end{CD} The point is that the composition of such classes may not be well-defined. However, when $S$ is a multiplicative system in $C$, we can readily show that the composition is well-defined in the sense that, when composing $[u_1]$, $[v_2]$, it is independent of the choice of isomorphism $y_{21}$ as well as the choices of the other 4 arrows fitting into the following commutative diagram, where vertical arrows are isomorphism: $\require{AMScd}$ \begin{CD} X_1 @>u_1>> Y_1 @>v_1>>Z_1\\ @A x_{12} AA @AA y_{21} A @AAz_{21}A\\ X_2 @>>u_2> Y_2 @>v_2>>Z_2. \end{CD} One can also show that the identity and associativity axioms hold, with the obvious identities $1_{[X]}$ as the class containing $1_X$ for each $X\in C$. Thus, we essentially have a category $[C]$ when isomorphisms in $C$ form a multiplicative system.

Here I follow the notion in https://stacks.math.columbia.edu/tag/04VB .

Moreover, when $C$ is preadditive, $[C]$ is preadditive with the induced additive structure, where we sum up two classes by choosing the same representative of source and target object.

Am I missing anything in the above argument? Is this really a thing? since I don't know how to search for similar results, except for this MO post by my friend:

Why “modding out the homeomorphism” in the category Top makes no rigorous sense?

Anyway, if the above is all correct, we actually yield a functor $$[\;]:Fun\rightarrow Fun,\quad C\mapsto [C]$$ Here I have another question:

Is $\operatorname{Fun}_+$ where objects are preadditive categories, with additive functors as morphisms, a category? If so, $[\;]$ can be restricted to $\operatorname{Fun}_+$.


(1) Should it be correct, another problem is that modding out too much objects and morphisms can wipe out many properties and structure of a category. So I wonder if there's other similar way of introducing new category, given a category.

(2) When a multiplicative system $S$ in $C$ is saturated (composable $f$, $g$, $h$ s.t. $ fg,gh \in S $ implies $g \in S$), then $S$ must contain all the isomorphisms of $C$. This condition seems strong to me, and I wonder if there is weaker condition to ensure that isomorphisms in a category form a multiplicative system.

  • 1
    $\begingroup$ It's unclear what the question is. Usually mathoverflow is not to check if an idea is correct. Also, I don't know what is meant by "multiplicative system" in a category. For localization of categories have you look at the book by Gabriel and Zisman? $\endgroup$ Jun 20 '19 at 21:48
  • $\begingroup$ @DavidWhite by ''multiplicative system'' I mean the same thing as in Gabriel and Zisman's book, when S admits a calculus of left and right fractions. Sorry for not clarifying my question. So my main questions are now in the remark part. I think this process of modding out morphisms is a bit different from localization of a category, but I will check in GZ's book later! Thank you. $\endgroup$ Jun 21 '19 at 6:08
  • 1
    $\begingroup$ You should almost never make a new category by collapsing objects: you should add in isomorphisms instead. If this means inverting a bunch of morphisms like Gabriel–Zisman, then that's fine. But you might be thinking of forming the skeleton of a category, which is not a quotient, but a subcategory. $\endgroup$ Jun 21 '19 at 6:35

The class of isomorphisms in a category is always a multiplicative system. But that doesn't help, as far as I can see, in showing that the composition in your putative category $[C]$ is well-defined. If $f_1 \sim f_2$ via $x_{12}$ and $y_{12}$, and $g_1 \sim g_2$ via $y_{12}'$ and $z_{12}$, then since $y_{12}\neq y_{12}'$ I don't see any way to show that $g_1 \circ f_1 \sim g_2 \circ f_2$.

This is actually one of the subtlest things about category theory: you can't generally "quotient by isomorphisms" to get rid of them. As David Roberts said in a comment, if you want to replace a category by one in which any two isomorphic objects are equal, you generally have to do it by choosing a skeleton rather than by passing to any sort of quotient.


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.