Switch thermals
Conduction loss at the hot R_DS(on). Working from the cold figure understates it several times over: for the default device the two differ by 2.6×. The check is the one davekni uses: what thermal resistance you are allowed to spend, rather than guessing at a heatsink.
whole bridge340 Waveraged over 14 % duty, 8 devices
per device, in burst304 W40.0 A RMS at the hot R of 0.19 Ω
transient Z_th0.329 K/Wover 2000 µs for ΔT 100 K
I_RMS per position = I_pk/2 (it conducts half a sine) per device = /n P_cond = I_RMS² · R P_sw = E_off · f allowed Z_th(pulse) = ΔT / P_burst allowed R_th(steady) = ΔT / P_average
Switching loss is entered by hand, because with ZVS working it is close to nothing and when ZVS breaks down it goes through the roof, and there is no predicting it. Leave it at 0 to see the conduction loss on its own, and read that as the floor: if ZVS fails, this margin is gone. It is another reason to set the phase lead generously.
The allowed Z_th must be checked against the transient thermal impedance curve in the datasheet at your pulse width, not against the static R_th(j-c).
The ΔT field is the largest free input here, and the two published answers are a factor of three apart because they subtract from different junction targets. It defaults to 100 K, which is davekni's, a 150 °C die over a 50 °C heatsink. Mads Barnkob keeps the ripple between junction and case under 30 K instead, from a junction wanted under 80 °C over a case at 50, and says his reason is second hand bricks of unknown history. Barnkob also settles the choice himself: it comes down to what temperature rise you find acceptable for your application. Which one you type is a decision about how much life you are buying, and a pulse rating is a temperature is the page about it.
davekni's rule of thumb for the order of magnitude: devices fail at about 4× the continuous or 2× the pulsed rated current.