Suppose we have two functors $F:C\leftrightarrow D:G$ and a morphism $\varepsilon:FG\rightarrow\operatorname{Id}_D$. I am looking for a way to check whether $\varepsilon$ is the counit of an adjunction. Mostly I am interested in a way to show that $\varepsilon$ is not a counit, i.e. necessary conditions.

  • 3
    $\begingroup$ A tautological condition is that the morphism $Hom(A,GB) \stackrel{F}\to Hom(FA,FGB) \stackrel{\varepsilon}\to Hom(FA,B)$ is an isomorphism for all $A \in C$, $B \in D$. $\endgroup$
    – Sasha
    Commented Jan 27, 2014 at 23:28
  • $\begingroup$ There are conditions on $F, G$, of course, e.g. that $F$ preserves colimits and $G$ preserves limits. Are you in a situation where you suspect $F, G$ may be part of an adjunction but you aren't sure what the counit should be? $\endgroup$ Commented Jan 28, 2014 at 1:17
  • $\begingroup$ You meant unit? Yes, I have a candidate for a co-unit but no explicit unit, and I need to show that sometimes my candidate cannot be a co-unit $\endgroup$
    – Adam Gal
    Commented Jan 28, 2014 at 9:15

1 Answer 1


A morphism $\varepsilon\colon FG\to Id_D$ is the counit of adjunction $F\dashv G$ iff for every $d\in D$ the morphism $\varepsilon_d\colon FGd\to d$ is universal from $F$ to $d$(i.e. is the terminal object of the comma category $(F\downarrow d)$). So we have a family of necessary conditions, parametrized by objects of $D$.

Also, there are interesting connections between $G$ and $\varepsilon$(if $\varepsilon$ is the counit):

  • $G$ is faithful iff for any $d\in D$ the morphism $\varepsilon_d$ is an epimorphism;
  • $G$ is full iff for any $d\in D$ the morphism $\varepsilon_d$ is a split monomorphism.

So, for example, if $G$ is full, and for some $d_0\in D$ the morphism $\varepsilon_{d_0}\colon FGd_0\to d_0$ does not have a left inverse, then $\varepsilon$ is not a counit.

Finally, another interesting property you may find useful:

  • If one of the functors $F$, $G$ is full, then the natural transformation $G\varepsilon$ is invertible.


Maclane, CFWM, chapter IV.

Edit: As you said, your $\varepsilon$ is invertible. Assuming $\varepsilon$ is the counit, you can also define the unit of the corresponding adjunction on the image of $G$ on objects(it follows from the triangular identities): $$ \eta_{G(d)}=(G(\varepsilon_d))^{-1} $$ Maybe it will help you to find the whole unit morphism. For example, if $G$ is surjective on objects, you don't need to check anything(in this case $(F,G)$ is an equivalence).

Also you can use some additional data about your categories to find the adjunction. For instance, if $D$ is locally small, complete, well-powered, has a small cogenerating family, and $G$ is continuous, then you can find the left adjoint by the Special Adjoint Functor Theorem. Then you can try to prove that $F$ is naturally isomorphic to this left adjoint etc.


Borceux, "Handbook of categorical algebra", vol.1.

  • $\begingroup$ I looked at my morphism, and sadly this doesn't help me since it is given as an inverse, so is invertible... $\endgroup$
    – Adam Gal
    Commented Jan 28, 2014 at 11:49
  • $\begingroup$ @AdamGal Is your $G$ fully faithful? Also, have you checked, that every $\varepsilon_d$ is a terminal object in $(F\downarrow d)$(it was the main part of my answer)? Anyway, this is a vague question as long as you don't mention your concrete $F$, $G$ and $\varepsilon$. $\endgroup$
    – Oskar
    Commented Jan 28, 2014 at 12:56
  • $\begingroup$ I am trying to check if it is terminal, but this is not so straitforward - amounts to showing that some map doesn't exist or is not unique, so there is a lot of guesswork. I am waiting to see if maybe someone knows something more. $\endgroup$
    – Adam Gal
    Commented Jan 28, 2014 at 14:00
  • $\begingroup$ @AdamGal See also edit. $\endgroup$
    – Oskar
    Commented Jan 28, 2014 at 14:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.