I just learned from Wikipedia that coproduct of two (commutative) rings is given by tensor product over integers, and that coproduct of a family of rings is given by a "construction analogous to the free product of groups." Can the tensor product approach be generalized to an arbitrary family of rings? (Infinite tensor product?) I'm a little surprised that coproduct of commutative rings requires noncommutative structure (free group). Does someone have a reference which explicitly constructs the coproduct?

  • $\begingroup$ Your first sentence uses the word coproduct one more time that you intended, I think. $\endgroup$ – Mariano Suárez-Álvarez Sep 17 '12 at 1:22

Your question is mildly ambiguous, so let's separate the statements for the categories of commutative rings and rings.

The infinite coproduct of commutative rings in $\text{CRing}$ is not given by an infinite tensor product, the reason being intuitively that you can only multiply finitely many elements of a ring at a time and infinite tensor products deal with attempting to multiply infinitely many things at a time. It is in fact given by a filtered colimit of finite tensor products. More precisely, let $A_i, i \in I$ be an indexed family of commutative rings, and for every finite subset $J$ of $I$ consider the tensor product $A_J = \bigotimes_{i \in J} A_i$. Then the infinite coproduct $\bigsqcup_i A_i$ is the filtered colimit of the $A_J$ equipped with all inclusion maps $A_{J_1} \to A_{J_2}$, where $J_1 \subset J_2$. The inclusion maps look like this:

$$a_{i_1} \otimes ... \otimes a_{i_n} \to a_{i_1} \otimes ... \otimes a_{i_n} \otimes 1_{A_{i_{n+1}}} \otimes ... \otimes 1_{A_{i_m}}$$

where $|J_1| = n$ and $|J_2| = m$.

The infinite coproduct of rings in $\text{Ring}$ is an infinite free product. As an abelian group, it is constructed similarly to the above; for an indexed family $A_i, i \in I$ of rings it consists of a filtered colimit (of abelian groups!) of finite tensor products (as abelian groups!) of the $A_i$ where we can repeat factors (due to the fact that we cannot assume that the images of $A_i$ and $A_j$ commute in general). The multiplication is given by concatenation. I don't know a reference for the construction but it consists more or less in following one's nose. When in doubt, always refer back to the universal property.

In particular, the functor $\text{CRing} \to \text{Ring}$ does not preserve coproducts; the phrase "coproduct of commutative rings" has a different meaning depending on whether you want to take the coproduct in $\text{CRing}$ or in $\text{Ring}$ because the corresponding universal properties are different.

  • 3
    $\begingroup$ It's worth noting that any category with filtered colimits and finite coproducts has all small coproducts, by the construction you give. $\endgroup$ – Zhen Lin Sep 17 '12 at 2:34
  • 1
    $\begingroup$ @Zhen: the construction I give in $\text{Ring}$ is not quite a construction using finite coproducts; the tensor products I'm considering are not free products. I'm just trying to describe the free product as an abelian group. I should make this clearer. $\endgroup$ – Qiaochu Yuan Sep 17 '12 at 2:47
  • $\begingroup$ Thanks for your helpful answer. For the filtered colimit approach, do you think one needs to order $I$ or mod out the finite tensor products to isomorphism classes or something, in order to have well-defined maps $A_{J_1}\rightarrow A_{J_2}$ ? $\endgroup$ – ashpool Sep 17 '12 at 14:49
  • 2
    $\begingroup$ @ashpool: not for $\text{CRing}$. The maps come from inclusions of finite subsets. That definition makes sense without an order. I wrote down the terms in order for simplicity but that's not necessary in $\text{CRing}$ because the factors in a tensor product commute. For the construction in $\text{Ring}$ you need to take all possible orders. $\endgroup$ – Qiaochu Yuan Sep 17 '12 at 16:28
  • 1
    $\begingroup$ @ashpool: actually the colimit is the same either way; that is, I could either only have $A_1 \otimes A_2$ or I could have both, and as long as the inclusion maps are what they should be it doesn't matter. The colimit still has the same universal property. $\endgroup$ – Qiaochu Yuan Sep 17 '12 at 18:07

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.