4
$\begingroup$

I'm trying to find the real nature of the so-called "projection arrow" for $f^* \dashv f_* $: $$ f_*(A\otimes f^*B)\leftarrow f_*A\otimes B $$ which can be found in any monoidal category $(\mathbf V,\otimes)$, and sometimes is an isomorphism (for example, when you look at categories of sheaves and there are suitable hypoteses of flatness over $A,B$).

My problem is that the machinery of "THC-situations" proposed in Gray's paper "Closed categories, lax limits and homotopy limits" seems to apply... backwards, and I'm not able to reverse it.

Let me recall a couple of things:

  1. A THC situation consists of three categories $A_1,A_2,A_3$ and functors $T\colon A_1\times A_2\to A_3$, $H\colon A_2^o\times A_3\to A_1$, $C\colon A_1^o\times A_3\to A_2$ such that $$ A_3(T(a_1,a_2),a_3)\cong A_1(a_1,H(a_2,a_3))\cong A_2(a_2, C(a_1,a_3)) $$ $T,H,C$ have respectively to be regarded as generalized tensor, hom and cotensor operations (a paradigmatic example of a THC situation is that of a monoidal closed category, where $T=-\otimes-, HC=[-,-]$). One can denote a THC situation as $\mathfrak t = \lbrace A_1,A_2,A_3,\begin{pmatrix}T\\ H\\ C\end{pmatrix} \rbrace$.
  2. One is now interested in linking two THC situations $\mathfrak t, \mathfrak t'$ via adjoint pairs. In what follows $\check G_i \dashv G_i\dashv \hat G_i$ is a string of adjoint functors where $G_i\colon A_i\to A_i'$, for $i=1,2,3$. Now it's easy to prove, for example, that $$ G_3T(\check G_1-,\check G_2-),\;\; G_1H(\check G_2-,\hat G_3-),\;\ G_2C(\check G_1-,\hat G_3-) $$ is a THC situation.
  3. It's also easy to prove that $G_1H(a_2,a_3)\cong H'(G_2a_2,G_3 a_3)$ if and only if $T(\check G_1 a_1',a_2)\cong \check G_3T'(a_1',G_2a_2)$, by simple adjoint calculus and the Yoneda Lemma: on the one hand $[a_1', G_1H(a_2,a_3)] \cong [\check G_1 a_1', H(a_2,a_3)] \cong [T(\check G_1a_1', a_2), a_3]$, and on the other hand $[a_1',H'(G_2a_2, G_3a_3)] \cong [T'(a_1',G_2a_2),G_3a_3] \cong [\check G_3T'(a_1',G_2a_2),a_3]$.

This should have now the following pleasant corollary: if I am given two monoidal closed categories $(\mathbf V,\otimes, [-,=])$, $(\mathbf V',\otimes', [-,=]')$ and a couple of adjoint functors $F\colon \mathbf V\dashv \mathbf V'\colon G$, then $G[V,W]\cong [GV,GW]'$ iff $FV'\otimes W\cong F(V'\otimes' GW)$, which is exactly what I need... backwards, because what I need is that the right adjoint behaves like this: $$ GV'\otimes W\cong G(V'\otimes' FW) $$ in order to justify the projection formula. Is there a similar TH(C) situation where I can get what I want?

$\endgroup$
2
  • 8
    $\begingroup$ Asking about THC situations when one's called tetrapharmakon.. really! ;) $\endgroup$ Jun 3, 2013 at 16:22
  • 2
    $\begingroup$ I strive for coherence. ;) $\endgroup$
    – fosco
    Jun 3, 2013 at 18:39

1 Answer 1

2
$\begingroup$

Here are some references that you may find helpful.

  1. The most common modern term for "THC-situation" is a two-variable adjunction. They can indeed be composed with ordinary adjunctions as you suggest. More generally they are the binary morphisms in a multicategory, and indeed a polycategory, whose unary morphisms are ordinary adjunctions.

  2. The equivalence between a right adjoint being closed monoidal and a "Hopf/Frobenius" condition relating it to its left adjoint is also standard. This is sometimes called "Frobenius reciprocity" and other times a Hopf adjunction.

  3. The "projection formula" map is indeed "backwards" from the Hopf adjunction map. The only abstract study I know of that mentions it is Fausk-Hu-May. Their equation (3.6) gives a general way to define such a morphism for any lax monoidal right adjoint, but they don't mention any "mate" property for it analogous to the Hopf-adjunction/closed-functor duality, and in general it seems such formulas are usually only isomorphisms for a restricted class of "dualizable" objects.

$\endgroup$

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.