Would you please clarify your question? Your namelist.input indicates that is a triply-nested case with very high resolution of 0.388km in your finest domain. When you look at GRDFLX, did you find extremely large negative values over all the 3 domains? What did you mean by ' simulate 100% PV'?
I also have a few concerns about your namelist options:
(1) cumulus scheme should be turned off for all the 3 domains
(2) The basis assumption of PBL scheme becomes invalid when the resolution is smaller than 1km. This will cause some trouble in your D03.
(3) why do you set gwd_opt = 1, 0, 1 ? I would suggest that you turn off GWD over all domains.