When you say "Dc analysis fails" it could be that only DC Newton failed, and some pseudo-transient method succeeded (where p-t starts at an "easy" solution, like zero voltages, zero currents).

It's also possible (and some "fast spice" simulators are known to do this) to just start the transient without a known dc operating point.  Any vector of voltages (and branch currents) can be used, and the simulator just proceeds.  Consider if you stopped the transient in the middle of a simulation, then took that operating point as the starting point for a new transient. You have to start up with Backward-Euler so your integration formula doesn't need more data, but there's no requirement that the operating point be a dc solution.