The primary limitations will be the precision of the numbers used in the solver (ie 32bit 64bit, and then double, double double, quad double) and the excitations. For example, in HFSS you can take the ports down to about 5kHz and have very accurate responses in HFSS13 as they have nicely improved this. However, if you transition from ports to voltage or current sources, you can go down to 10s of Hz and have excellent field results. It is also of note that most structures in these frequencies are horrid radiators, so your terminating boundary (Radiation boundaries, PMLs, etc) can be brought MUCH closer than lambda/4 or lambda/10 as these structures have most of the fields in the reactive near field. Just food for thought!
Have Fun!