TITLE: Dualizable presheaves with respect to Day convolution
QUESTION [6 upvotes]: This question was posted on MSE and got very little attention, so I'm also posting it here.
Let $\mathcal{C}$ be a closed symmetric monoidal category and let $PSh(\mathcal{C}):=Fun(\mathcal{C}^{op}, \mathbf{Set})$ be its category of presheaves regarded as a closed symmetric monoidal category via Day convolution of presheaves.
Is there a nice description of the dualizable objects of $PSh(\mathcal{C})$
in terms of the dualizable objects of $\mathcal{C}$? For example, could it be that the dualizable presheaves $PSh(\mathcal{C})_{fd}$ consists of objects given as filtered (co)limits of dualizable objects in $\mathcal{C}$?

REPLY [6 votes]: Lemma: In a closed symmetric monoidal category where the unit object $1$ is tiny (meaning $\text{Hom}(1, -)$ preserves colimits), every dualizable object is tiny.

Proof. If $x$ is dualizable, then $\text{Hom}(x, -) \cong \text{Hom}(1, x^{\ast} \otimes (-))$. Since the monoidal structure is assumed to be closed, $x^{\ast} \otimes (-)$ preserves colimits, and by assumption so does $\text{Hom}(1, -)$. $\Box$
The unit object in $\text{Psh}(C)$ is the presheaf represented by the unit object; since colimits in presheaf categories are computed pointwise, every representable presheaf is tiny, so the lemma applies and we conclude that every dualizable object in $\text{Psh}(C)$ is tiny. But it's standard that the tiny objects in $\text{Psh}(C)$ are precisely the retracts of the representable presheaves. Among these, the dualizable objects include the presheaves represented by dualizable objects in $C$, as well as (I think?) their retracts. I don't know if it's possible for a nontrivial retract of a presheaf represented by a non-dualizable object to be dualizable.