5
$\begingroup$

Recently I was having an algebra lecture that introduces category theory and encountered a problem about the nonexistense of the right adjoint functor of the forgetful functor $U\colon\mathbf{Ab}\to\mathbf{Grp}$. I know it admits a left adjoint, which is defined by the abelianization functor $\mathbf{Grp}\to\mathbf{Ab}$. However it is not so obvious to me that $U$ does not have a right adjoint.

What I think may be useful is a proposition that $F\colon\mathcal C\to\mathcal D$ has a right adjoint if and only if the functor $\mathrm{Hom}(F-,Y)$ is representable for each $Y\in\mathrm{Ob}(\mathcal D)$ but it seems a bit difficult to show a functor not representable, nor nonexistence of natural transformations between two functors. So I would like to ask for some explanation of the nonexistence of right adjoint of such a forgetful functor, and thanks in advance...

$\endgroup$
13
$\begingroup$

Hint: Left adjoints preserve colimits, so it suffices to prove that $U$ does not preserve colimits.

For an even bigger hint, hover over the box below.

$U$ doesn't even preserve binary coproducts. The coproduct of two groups in $\mathbf{Ab}$ is their direct sum, but their coproduct in $\mathbf{Grp}$ is their free product.

$\endgroup$
  • $\begingroup$ Thank you for your help! $\endgroup$ – josephz Mar 19 '18 at 16:49
4
$\begingroup$

Clive gave a neat short way to answer your question. But let me try to explain what is really going on under the hood here.

Call a Lawvere theory any bijective-on-objects finite-coproducts-preserving functor $j : \aleph_0 \to \mathcal L$ where $\aleph_0$ is a skeleton of the category of finite sets. Objects of $\aleph_0$ (and so also those of $\mathcal L$) are just the integers $0,1,2,\dots$ and the morphisms $n \to m$ in $\aleph_0$ are the set-maps from $\{0,1,\dots, n-1\}$ to $\{0,1,\dots,m-1\}$. A model of the theory $j$ is a presheaf $M : \mathcal L^{\rm op} \to \mathbf{Set}$ that preserves finite products. The category of models of $j$ is the full subcategory of $\widehat{\mathcal L}$ spanned by the model of $j$; let's denote it $\mathbf{Mod}_j$. You might want to think of the morphisms in $\mathcal L(1,n)$ as $n$-ary operations and commutative diagrams in $\mathcal L$ as axioms; then the models are to be thought as sets equipped with these operations and satisfying the axioms.

This might seem a strange notion, but you actually know at least two (and probably many more) Lawvere theories : the theory of groups and the theory of abelian groups. The first one is described by the category $\mathcal L_{\rm gr}$ given as the full subcategory of $\mathbf{Grp}$ spanned by the free groups on finite sets; the fuctor $j_{\rm gr}$ is the obvious one, mapping $n$ to the free group on $\{0,1,\dots,n-1\}$. The second one is described by the category $\mathcal L_{\rm ab}$ given as the full subcategory of $\mathbf{Ab}$ spanned by the free abelian groups on finite sets; the fuctor $j_{\rm ab}$ is the obvious one, mapping $n$ to the free abelian group on $\{0,1,\dots,n-1\}$. If you are careful enough, you can check that $\mathbf{Grp}\simeq \mathbf{Mod}_{j_{\rm grp}}$ and $\mathbf{Ab}\simeq \mathbf{Mod}_{j_{\rm ab}}$.

Of course, we want a way to compare those theories. There is a very natural notion of morphisms between theories $j_1$ and $j_2$: these are the functors $\varphi$ such that $\varphi j_1 = j_2$. In the case $j_1 = j_{\rm grp}$ and $j_2 = j_{\rm ab}$ there is such a morphism $\varphi$: it maps the free groups on $n$ elements to the free abelian groups on $n$ elements; the action of maps takes advantages of the fact that an element of the free groups on $n$ elements can equally be seen as an element of the free abelian group on $n$ elements (just not normalized). More hand-waving, this morphisms of theory reflects the fact that axioms of groups are (some) axioms of abelian groups. Going back to a general $\varphi$, it gives a restriction functors: $$ \varphi^\ast : \widehat{\mathcal L_2} \to \widehat{\mathcal L_1} $$ By the theory of Kan extensions, this functor admits a left and a right adjoint, usually denoted $\varphi_!$ and $\varphi_\ast$ respectively. Now remark that $\varphi^\ast$ maps finite-products-preserving presheaves to finite-products-preserving presheaves, so that it induces a functor $$ U_\varphi : \mathbf{Mod}_{j_2} \to \mathbf{Mod}_{j_1} $$ This functor will have a left adjoint precisely if $\varphi_!$ maps finite-products-preserving presheaves to finite-products-preserving presheaves and it will have a right djoint precisely if $\varphi_\ast$ maps finite-products-preserving presheaves to finite-products-preserving presheaves. Here is the catch: $\varphi_!$ always does while $\varphi_\ast$ has no reason to! (It has to do with the representation of presheaves as canonical colimits of representables, but I dont want to burry you under details here.)

In the case $j_1 = j_{\rm grp}$ and $j_2=j_{\rm ab}$ and the morphism $\varphi$ sketched earlier, $U_\varphi$ is exactly the forgetful functor from abelian groups to groups, which admits a left adjoint (the abelianization functor) but have no reason to admit a right adjoint.


As you probably guessed, this framework applies to most algebraic theories you encountered (more precisely, the unisorted first-order equational theories). I like the stratospheric view that this abstract point of view gives you: the real reason behind the non existence of a right adjoint in your case is that "abelianity" is an additional axiom on the usual structure of group.

$\endgroup$
  • $\begingroup$ Wow, thank you!! Your profound explanation is much appreciated! $\endgroup$ – josephz Mar 20 '18 at 15:14

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.