Diagonal of connectivities for diffusion maps is zero - is this intended?

My question is about the connectivities used within this function: https://github.com/scverse/scanpy/blob/0692ef9ea30335b95f7e7f9aab7be856469d9f35/scanpy/tools/_dpt.py#L16
the computation that uses connectivities is as follows:
https://github.com/scverse/scanpy/blob/0692ef9ea30335b95f7e7f9aab7be856469d9f35/scanpy/neighbors/__init__.py#L914-L925

I can follow most of this computation and link it back to the main reference, except for the fact that the connectivities (i.e. adata.obsp['connectivities']), when calculated using scanpy.pp.neighbors as suggested in the diffmap docstring, have a zero diagonal.

Could someone confirm whether this is the intended calculation? And if so, provide a reference that confirms this? I've not read the reference in detail, but I would've thought (following section 3.1 and 5) that, as all the computations revolve around the usage of a kernel, the connectivities should be positive definite before normalization, which wouldn't be the case if the diagonal was zeroed out. At a glance, I also cannot find anywhere in that reference that talks about P(xi, xi).

Read the linked sections of scanpy/tools/_dpt.py and scanpy/neighbors/init.py, then compare the connectivities construction with sections 3.1 and 5 of the cited diffusion maps reference. The issue is complete when the intended diagonal behavior is confirmed and the repository has a clear code or documentation follow-up if needed.