23

Let $C$ and $D$ be categories and $F:C\to D$, $G:D\to C$ two functors.

$F$ is left-adjoint to $G$, if there are natural transformations $\eta:id_C\to GF$ and $\epsilon:FG\to id_D$ such that \begin{eqnarray} F&\xrightarrow{F\eta}&FGF&\xrightarrow{\epsilon F}F\\ G&\xrightarrow{\eta G}&GFG&\xrightarrow{G\epsilon}G \end{eqnarray} are the identity (!) transformations.

$F$ is an equivalence of categories (with inverse $G$) if there are natural isomorphisms $\eta:id_C\to GF$ and $\epsilon:FG\to id_D$ without any further properties.

Is $F$ left-adjoint to $G$, if $F$ is an equivalence of categories (with inverse $G$)?
If not, suppose that $F$ is an equivalence of categories with inverse $G':D\to C$ and suppose further that $F$ is left-adjoint to $G$. Does it follow that there is a natural isomorphism $G\to G'$ or is there even an identity $G=G'$?

Arnaud D.
  • 20,884
sopot
  • 555
  • 3
  • 9
  • The answer to the first question is "yes", $F$ is both left-adjoint and right-adjoint to $G$. The answer to the second question is "yes, there exists a natural isomorphism $G\to G'$, but it is not necessarily an identity". – Oskar Dec 06 '13 at 14:47

1 Answers1

21

Every equivalence can be "improved" to an adjoint equivalence, but you may have to change the unit or the counit: see e.g. here for a proof. Essentially the same thing is proved as Theorem 4.2.3 in the Homotopy Type Theory book.

Zhen Lin
  • 90,111
  • What is the Homotopy Type Theory book? – Martin Brandenburg Dec 06 '13 at 16:35
  • 50
    @MartinBrandenburg: it can be characterized as either the unique book with title "Homotopy Type Theory" or as the unique book on the subject of Homotopy Type Theory. – Omar Antolín-Camarena Dec 06 '13 at 20:55
  • 4
    http://homotopytypetheory.org/book/ – Berci Dec 08 '13 at 00:42
  • Thanks, Zhen Lin. Does it follow that there is a natural isomorphism $G\cong G'$? – sopot Dec 08 '13 at 17:27
  • 1
    Adjoints are unique up to unique isomorphism. – Zhen Lin Dec 08 '13 at 17:53
  • 1
    Suppose we have already shown that the natural iso $\eta : 1_{\mathbf{C}} \to GF$ witnessing the equivalence is the unit of the desired adjunction. Why can't the counit just be the other iso $\epsilon : FG \to 1_{\mathbf{D}}$ of the equivalence in general? It satisfies the UMP the counit since for each $g: FC \to D$, we have $\epsilon_D \circ (\epsilon_D^{-1} \circ g) = g$ and $\epsilon_D^{-1} \circ g = Ff$ for some unique $f : C \to GD$, since we can show that $F$ is fully faithful. What is wrong with this argument? – StudentType Jan 28 '16 at 07:35
  • Yes, $\epsilon$ is a counit, but not necessarily for the same adjunction. – Zhen Lin Jan 28 '16 at 08:23
  • Could you develop a little bit more why this argument doesn't proof that it is a counit for the same adjunction? (Is it because it does not show that they are related by the triangle identities?) – StudentType Jan 28 '16 at 09:54
  • They would have to satisfy the triangle identities if they were unit and counit for the same adjunction. – Zhen Lin Jan 28 '16 at 09:57