tgpedersen wrote:
>>No, the a- is explained well enough by *h2-[*],
>
>
> Only if you assume *h2e-; if /a/ is preserved, rather than produced
> by h2 there's nothing wrong in *h2aN-.
But *a doesn't seem to have disappeared in the nil grade, while we have
clear reflexes of PIE *-h2p- in compounds such as Skt. dvi:pá- <
*dwi-h2p-ó-. Also, the strong cases in Indo-Iranian have *a: (Ved.
nom.pl. a:pas, Av. acc.sg. a:pm), which looks like a Brugmannian
lengthening of *o, so we should reconstruct *h2ó:p-s, nom.pl. *h2op-es,
acc.sg. *h2op-m., gen.sg. *h2ap-ós, with the same apophony as in
*pod-/*ped-. All this points to *h2a- < **h2e- in this word.
Piotr