# Clarifications on the definition of product in Category Theory

I am self-studying Hungerford's book Algebra. Here is the definition that I don't understand.

Let $\mathcal{C}$ be a category and $\{A_{i}|i\in I\}$ a family of objects of $\mathcal{C}.$ A product for the family $\{A_{i}|i\in I\}$ is an object $P$ of $\mathcal{C}$ together with a family of morphisms $\{\pi_{i}:P\rightarrow A_{i}|i\in I\}$ such that for any object $B$ and family of morphisms $\{\varphi_{i}:B\rightarrow A_{i}|i\in I\},$ there is a unique morphism $\varphi: B\rightarrow P$ such that $\pi_{i}\circ \varphi =\varphi_{i}$ for all $i\in I.$

Let me try to explain what I don't understand in the definition above. Every time I have to look back to see if the correct is "there is a unique morphism $\varphi: B\rightarrow P$" or "there is a unique morphism $\varphi: P\rightarrow B$".

After reading the comments bellow, I was able to understand the definition.

But I would like to know the motivation for this definition. I was wondering what lead someone to define "product" in a category?

And by using the Arturo's word (I hope with his permission) what is the intuition behind "products" in a category?

Besides, if a category has no product, what one can say about it? May I say that it is not a nice one?

• Think of examples. If you give me a set $Y$ and a bunch of maps $Y \to X_i$, then I naturally get a map $Y \to \prod X_i$. On the other hand, I don't see any way of turning this data into a map $\prod X_i \to Y$. – Dylan Moreland Feb 28 '12 at 0:21
• Dear spohreis: To follow up on Dylan, you can think of the the product of an empty family of sets, versus the disjoint union of such a family. – Pierre-Yves Gaillard Feb 28 '12 at 0:41
• Assuming that the product exists, and is constructed the way we're used to from categories of e.g. groups or topological spaces, the definition roughly reads "A morphism into the product of the family $\{A_{i}|i\in I\}$ is uniquely determined by its restriction to each coordinate." – Arthur Feb 28 '12 at 0:45
• Among all objects that can map onto every one of the objects $A_i$, the product is the one that contains no unnecessary "information" (it is the 'rightmost' object that maps onto all $A_i$, if you imagine your arrows going left to right). "Rightmost" objects (right universal objects) are the 'least general' ones; "leftmost" objects (left universal objects, such as free objects and coproducts) are the "most general" ones. – Arturo Magidin Feb 28 '12 at 5:13
• Don't take as an example a category of modules, because in such categories, finite coproducts and finite products are the same, so things are not very clear. The category of sets makes everything clear I think. – Lierre Feb 28 '12 at 22:36

The categorical definitions of the basic objects (products, coproducts, equalizers, coequalizers, limits, colimits, etc) arise from abstracting a particular situation that shows up in many instances.

The notion of a "product" shows up repeatedly among Sets (cartesian product), Groups (cartesian product again), Rings (cartesian product yet again), Modules (and again), Topological spaces (and again).

Their "object" properties vary from category to category; in the class of abelian groups, the direct product contains the direct sum, which is an important objects in its own right (it's also important in Groups, but less so). Among Unital Rings, however, the direct sum is not an interesting object when it is different from the product (because it is not a unital ring). In topological spaces, there is a "natural" way of defining the topology on a product (the "box" topology), but it turns out to be lacking in some uneasy sense when there are too many factors in the product... and there is an alternative possibility (the "product topology") that seems to be behave better...

The general, guiding principle in Category Theory, though, is that what is important is not what an object "is", but rather what it "does": how it behaves relative to other objects in terms of arrows/morphisms. So we want to abstract the properties that all those instances share in $\mathbf{Sets}$, $\mathbf{Groups}$, $\mathbf{AbGroups}$, $\mathbf{Rings}$, $\mathbf{UnitalRings}$, $\mathbf{Semigroups}$, $\mathbf{TopSpaces}$, etc. (In the latter case, it even suggests which of the two possible topologies "should" be used).

So... what do all these specific instances all have in common? From the cartesian product we can map to each factor: if we view the cartesian product $\times X_i$ as the set of functions $f\colon I\to \cup X_i$ with $f(i)\in X_i$ for each $i\in I$, then we have a natural map from the product to each $X_i$ by "evaluation": $\pi_i\colon f\longmapsto f(i)$. And $f$ is completely determined by these images; so that if you "select" $x_i\in X_i$ for each $i$, then this gives you a unique $\mathbf{x}\in \times X_i$.

Ah, but again: we don't want to think in terms of elements (what the objects "are"), but rather in terms of maps. For $\mathbf{Sets}$, selecting objects is the same as mapping from single element sets... but again we are worrying about the what the objects "are" (single element sets)...

Okay: since we care about maps, note this: not only can we go from $\times X_i$ to each $X_i$. These maps are such that if you have any set $Y$, and any maps $g_i\colon Y\to X_i$, then each $y\in Y$ determines an element of $\times X_i$, namely, $y\colon i\to g_i(y)$. Put another way:

If we let $\pi_i\colon\times X_i\to X_i$ be the "evaluate at $i$" map, then for any set $Y$ and any maps $g_i\colon Y\to X_i$, there is a way to map $g\colon Y \to \times X_i$ in such a way that $\pi_i\circ g(y) = g_i(y)$ for every $i$.

What's more, $g$ is completely forced by this condition. This is a good "categorical" property, because it is given entirely in terms of morphisms.

Does this property "go through" the other instances? The group with the coordinatewise product? The semigroup/ring/module with coordinatewise product? Yes! Good. What about topological spaces? Yes with respect to the product topology... no with respect to the Box topology... so we want to use the "product topology" for the "product".

This definition matches the many instances, and has nice "categorical" properties. It completely characterizes the product up to unique isomorphism. So it seems like a good definition. Turns out to be the right way to generalize the properties of the cartesian product/direct product to other categories.

• Do you have any links to more info about this "objects" vs "morphisms" mindset? I've read many articles and books on category theory (well, rather beginnings of books :Ь), but never seen that regard discussed, and hence didn't get why do we even care about commutative diagrams. – Hi-Angel Dec 25 '16 at 6:19