$\def\L{\mathfrak{L}}\def\Prof{\mathsf{Prof}}$ Recall that Isbell duality $\text{Spec}\dashv {\cal O} : {\cal V}^{A^°} \leftrightarrows \big({\cal V}^A\big)^°$ allows us to define the functor $$ \L : \Prof(A,B) \to \Prof(B,A) $$ ($\cal V$ is a cosmos in which $A,B$ are enriched categories) sending $K : A^° \times B \to \cal V$ into $\L(K) : (b,a)\mapsto {\cal V}^{A^°}(K(-,b), \hom(-,a)) = {\cal O}(K_b)_a$.

Is it true that $\L(H \circ K) \cong \L(K)\circ \L(H)$, maybe under some additional assumptions on $H,K$? Or maybe in general, but that's not a formal proof.

Is there any relation between $\text{Ran}_FK$, as a profunctor $C^° \times B\to \cal V$ and the composition $K\circ \L(F)$?

How does this construction interact with the "tautological dualiser" of $\Prof$ that sends $A \mapsto A^°$?

Is there any reference that encompasses all, or almost all, these questions?


This is the right Kan extension of $\hom_A: A \nrightarrow A$ along $K: A \nrightarrow B$ in the bicategory of profunctors. Which is to say that for every profunctor (aka bimodule) $L: B \nrightarrow A$ there is a natural bijection between morphisms $LK \to \hom_A$ and morphisms $L \to \mathcal{L}(K)$. Since $\hom_A$ is a unit $1_A$ in this bicategory, it may be more suggestive to write $\mathcal{L}(K) = Ran_K 1_A$.

Thus for $H: A \nrightarrow B$ and $K: B \nrightarrow C$, we have a canonical map $\mathcal{L}(K)K \to 1_B$, and whiskering on the right by $H$ and on the left by $\mathcal{L}(H)$, we get a composite

$$\mathcal{L}(H)\mathcal{L}(K)KH \to \mathcal{L}(H)1_B H \cong \mathcal{L}(H)H \to 1_A.$$

By the universal property of $\mathcal{L}(KH)$, we now obtain a canonical map $\mathcal{L}(H)\mathcal{L}(K) \to \mathcal{L}(KH)$.

But this map is typically not an isomorphism. The simplest type of example that comes to mind is just the classical case of bimodules over rings, where for a right-$B$ left-$C$ bimodule $K$, we have $\mathcal{L}(K) = \hom_A(K, B)$ regarded as a right-$C$ left-$B$ bimodule. We have in this situation a canonical map

$$\hom_A(H, A) \otimes_B \hom_B(K, B) \to \hom_A(K \otimes_B H, A)$$

but normally this won't be an isomorphism. Indeed, even in the humble case of vector spaces over a ground field $k$, the canonical map $V^\ast \otimes_k W^\ast \to (V \otimes W)^\ast$ isn't generally an isomorphism. Of course we do get an isomorphism here in some special cases, such as if $K$ is finitely generated projective over $B$. More abstractly, this is the situation where $K$ has a left adjoint bimodule, and here we may recall that in a 2-category or bicategory, if an arrow $K: B \nrightarrow C$ has a left adjoint $L$, then it is necessarily $L = Ran_K 1_B$, i.e., $L = \mathcal{L}(K)$ in our situation. In that case, the asserted inverse $\mathcal{L}(KH) \to \mathcal{L}(H)\mathcal{L}(K)$ of the canonical map is mated (by the adjunction $\mathcal{L}(K) \dashv K$) to an arrow $\mathcal{L}(KH)K \to \mathcal{L}(H)$, which in turn is mated to the canonical arrow $\mathcal{L}(KH)KH \to 1_A$ using the definition of right Kan extension.

[For the bicategory of profunctors or bimodules, such right adjoints $K$ are induced by functors $F: C \to \bar{B}$ where $\bar{B}$ is the Cauchy completion of $B$ (which we may think of as analogous to the category of finitely generated projective $B$-modules); specifically, $K(b, c) = \bar{B}(b, Fc)$ and its left adjoint is given by $\mathcal{L}(K)(c, b) = \bar{B}(Fc, b)$.]

By similar reasoning as above, one may calculate that if $K: B \nrightarrow C$ has a left adjoint $L$ (in a bicategory $\mathbf{B}$), then for every $F: B \nrightarrow A$ we have an isomorphism $Ran_F(K) \cong K \circ Ran_F(1_B)$, assuming these right Kan extensions exist. This is purely formal of course: for every $H: A \nrightarrow C$ we have isomorphisms

$$\mathbf{B}(A, C)(H, Ran_F(K)) \cong \mathbf{B}(B, C)(HF, K) \cong \mathbf{B}(B, B)(LHF, 1_B) \cong \mathbf{B}(A, B)(LH, Ran_F(1_B)) \cong \mathbf{B}(A, C)(H, K Ran_F(1_B)).$$

I'm not sure what could be said about interaction with the "tautological dualizer" (where now are viewing $\textbf{Prof}$ as a compact closed bicategory): essentially all of the above has to do with the bicategory structure, not the monoidal bicategory structure.

| cite | improve this answer | |
  • 2
    $\begingroup$ That's what I call an answer! $\endgroup$ – Fosco Jan 24 '17 at 8:06
  • $\begingroup$ Thanks! I'm not sure where in the literature this type of thing would be, but I'm sure it's there somewhere. $\endgroup$ – Todd Trimble Jan 24 '17 at 14:27
  • $\begingroup$ Any reference exhibiting a feeble relation with this construction is welcome! This definition is part of a particularly "concrete" (provided "concrete" is the right adjective here :-) ) construction when $\cal V = $chain complexes or, better, a simplicially co/tensored model category. $\endgroup$ – Fosco Jan 24 '17 at 14:51
  • $\begingroup$ On a separate note, I'm able to "blindly" prove that $L(H \diamond K)^a_c \cong Nat(H^c, L(K)^a) = Ran_H L(K)^a_c$ as a right Kan extension in $\bf Prof$. Your explanation trivializes this result. $\endgroup$ – Fosco Jan 24 '17 at 14:53

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.