I've encountered this exercise in Aluffi's Algebra: Chapter 0. It might be helpful to say that the book doesn't introduce functors at this stage,, and that the book defines a presentation of a group $G$ as a pair $(A\mid \Psi)$, where $\Psi$ is a subset of $F(A)$ such that $G \cong F(A)/R$, where $R$ is the smallest normal subgroup of $F(A)$ containing $\Psi$.

So, here is the exercise.

Let $(A\mid \Psi)$, resp., $(A'\mid \Psi')$, be a presentation for a group $G$, resp. $G'$; we may assume that $A,A'$ are disjoint. Prove that the group $G\ast G'$ presented by $(A\cup A'\mid \Psi\cup\Psi')$ satisfies the universal property for the coproduct of $G$ and $G'$ in $\mathsf{Grp}$.

(Use the universal properties of both free groups and quotients to construct natural homomorphisms $G\rightarrow G\ast G'$ and $G'\rightarrow G\ast G'$.

First of all, why would we assume that $A,A'$ are disjoint? If they are then their union is their disjoint union as well. But can any 2 groups be presented with $A\cap A' = \emptyset$?

Secondly, I still can't prove it even for disjoint $A$ and $A'$. So, will be thankful for any tips.

  • 2
    $\begingroup$ If $A$ and $A'$ are not disjoint, then $A\cup A'$ will not be the disjoint union of $A$ and $A'$. The presentation then does not yield the coproduct. (Or else $G * G$ would be $G$.) $\endgroup$ Jul 25, 2015 at 21:04
  • 1
    $\begingroup$ If you have any two groups $G$ and $G'$, you can always find two groups $H \cong G$ and $H' \cong G'$ such that $H \cap H' = \varnothing$. Thus, you get presentations with disjoint sets of generators. $\endgroup$ Jul 25, 2015 at 21:05
  • $\begingroup$ Thanks! Can't believe I didn't think of it. $\endgroup$
    – Jxt921
    Jul 25, 2015 at 21:35

4 Answers 4


Let $G\ast G'=(A \cup A' \mid \Psi \cup \Psi')$.

By the universal property for free groups, for the set map $f: A \to G\ast G'$ where $f(a)=a$, there exists a unique group homomorphism $\bar f: F(A) \to G\ast G'$ defined on the generating set $A$ by $\bar f(a)=f(a)=a$ and making the relevant diagram commute. Similarly, for $g: A' \to G\ast G'$, we have a homomorphism $\bar g: F(A') \to G\ast G'$.

Next, $R \subset \ker \bar f$ and so by the universal property for quotients there exists a unique homomorphism $\tilde f: G=F(A)/R \to G\ast G'$ defined by $\tilde f(aR)=\bar f(a)=f(a)=a$ and making the relevant diagram commute. Similarly, there exists a hom $\tilde g: G' = F(A')/R' \to G \ast G'$.

So, we actually have our natural maps $G \hookrightarrow G\ast G'$ and $G' \hookrightarrow G \ast G'$.

Now, for any group $Z$ and any pair of homomorphisms $\varphi: G \to Z$ and $\psi: G' \to Z$, we know these maps are determined by where they map the generators in $A$ and $A'$, respectively. So, there exists a unique map $\sigma: G\ast G' \to Z$ by mapping $a \in A$ to $\sigma(a)=\varphi(a)$ and $a' \in A'$ to $\sigma(a')=\psi(a')$.

  • $\begingroup$ How do you know that $\sigma$ is well defined? $\endgroup$
    – maroxe
    Dec 9, 2020 at 4:06
  • $\begingroup$ We are given that $A$ and $A'$ are disjoint and then $\sigma$ is well-defined since $\varphi$ and $\psi$ are. $\endgroup$
    – Juan L.
    Dec 9, 2020 at 4:32

A map out of a group with presentation $(A,\Psi)$ is just a choice of image for each element of $A$ that respects the relations $\Psi$. So a map out of $(A\sqcup A', \Psi\sqcup \Psi')$ is just a choice of images for each element of $A\sqcup A'$ respecting the relations in $\Psi$ and in $\Psi'$. But since the relations in $\Psi$ have nothing to do with $A'$ and vice versa, this is just the same as the data of a map out of $(A,\Psi)$ and a map out of $(A',\Psi')$.


Here's a simple case when $A$ and $A'$ aren't disjoint: let $A = A' = \{ x \}$, and let $\Psi = \Psi' = \varnothing$. Then $G=G'=F1$, that is, $G$ and $G'$ are both the free group on one generator.

The coproduct of $G$ and $G'$ in $\mathsf{Grp}$ is the free group on two generators, however $$(A \cup A' \mid \Psi \cup \Psi') = (\{x\} \mid \varnothing) = F_1$$ so that $(A \cup A' \mid \Psi \cup \Psi')$ is not the coproduct of $G$ and $G'$.

As for why $G * G' = (A \cup A' \mid \Psi \cup \Psi')$ when $A \cap A' = \varnothing$, prove that the inclusion maps $$G \hookrightarrow (A \cup A' \mid \Psi \cup \Psi'), \qquad G' \hookrightarrow (A \cup A' \mid \Psi \cup \Psi')$$ are homomorphisms and that if there is a group $H$ and homomorphisms $$f : G \to H, \qquad f' : G' \to H$$ then there is a unique homomorphism $\bar f : (A \cup A' \mid \Psi \cup \Psi') \to H$ such that $\bar f \upharpoonright G = f$ and $\bar f \upharpoonright G' = f'$. This is reasonably easy if you write down all the definitions, since $\bar f$ is determined by its action on the generators, i.e. the elements of $A \cup A'$.

  • $\begingroup$ Should we use the fact that $F(A\cup A′)$ is a coproduct of $F(A)$ and $F(B)$ in $\mathsf{Grp}$? Then we have a diagramm $\begin{array}{ccccc} F(A) & \stackrel {\iota_1} \to & F(A \cup A') & \stackrel{\iota_2} \leftarrow & F(A') \end{array}$ which gives us a unique homomorphism $F(A \cup A') \to G$ corresponding to all groups $G$ and pairs of homomorphisms $F(A),F(A') \to G$. $\endgroup$
    – Jxt921
    Jul 25, 2015 at 21:57
  • $\begingroup$ @Jxt921: I'm not sure what you mean. Knowing that $F(A \cup A')$ is a coproduct of $F(A)$ and $F(A')$ in $\mathsf{Grp}$ only helps you in the cases when $\Psi=\Psi'=\varnothing$. To prove that $(A \cup A' \mid \Psi \cup \Psi')$ is a coproduct in $\mathsf{Grp}$, quite literally define the map $\bar f$ (given $f,f'$) as I suggested in my answer. More specifically, define $\bar f(a) = f(a)$ if $a \in A$ and $\bar f(a) = f'(a)$ if $a \in A'$. Prove that this extends to a unique homomorphism $(A \cup A' \mid \Psi \cup \Psi) \to H$, which extends both $f$ and $f'$. $\endgroup$ Jul 25, 2015 at 22:50
  • $\begingroup$ I'm not sure I understand the relation between $R,R'$ and $R''$, where $G \cong F(A)/R$, $G' \cong F(A')/R'$ and $G*G' \cong F(A \cup A')/R''$. We know only that $R''$ is the smallest normal subgroup of $F(A \cup A')$ containing $\Psi \cup \Psi'$. What does it tell us about $R''$? $\endgroup$
    – Jxt921
    Jul 26, 2015 at 19:52

Here's what I tried to do.

Let's construct a homomorphism $F(A) \xrightarrow{\rho} F(A\cup A')/R''$, where $R''$ is the normal closure of $\Psi \cup \Psi'$. Define $\rho(w) = wR''$. Then by universal property of quotients(and since the normal closure of $\Psi$ $R \subseteq R''$) we have a unique homomorphism $F(A)/R \xrightarrow{i_1} F(A \cup A')/R''$ defined by $i_1(wR) = wR''$.

Similarly, we have a homomorphism $F(A')/R' \xrightarrow{i_2} F(A \cup A')/R''$.

Then for all groups $G$ and homomorphisms $F(A)/R \xrightarrow{f} G$, $F(A')/R' \xrightarrow{g} G$ we have aunique correspoding morphism $F(A \cup A')/R'' \xrightarrow{\phi} G$ defined $\phi(wR'') = f(wR)$ if $w \in F(A)$ and $\phi(wR'') = g(wR')$ if $w \in F(A')$. If $w$ is a word which consist of letters from both $A$ and $A'$ then(since \phi is a homomorphism) we have $\phi(w) = ...f(a_1R)...g(a'_1R')...$ and so on.

Then $F(A \cup A')/R''$ satisfies a universal property of a coproduct of $F(A)/R$ and $F(A')/R'$.


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.