Dear ION,
I would like to add some more details about this experiment.
The idea is quite simple, we have a gas (or mix of gases) and use step-up flyback to create very sharp high voltage pulses (electric field).
Gas ionized and avalanche discharge occurs.
Unlike in conductors (wires) here we get here two different types of charges carriers, light electrons and heavy ions. So charge carriers move in different directions and with significantly different speeds. If we manage make our HV pulse short enough current is not closed in secondary and power source used only to trigger avalanche.
We "collect" electrons into capacitor and slow ions capture electrons from surrounding or from spark gap electrodes and became again neutral.
This is very simplified description of processes here.
We can repeat this process and each time get some charge into capacitor.
Now about power calculations. We know power supply voltage and pulse duration. Current thru primary grows linearly, so measuring maximum current (at the end of pulse) allows us calculate energy spend on one pulse E = U poser supply * T pulse * I max / 2
We also can calculate energy in capacitor observing voltage on it E = C U ^ 2 / 2
Interesting that we spend same amount of energy to start process, so with n pulses spent energy grows linear, but energy stored in capacitor grows quadratically.
in other word it is inevitable that at some point stored energy in capacitor will exceed energy spent to trigger avalanches.
Maximum voltage to which capacitor can be charged with such series of pulses depends on HV pulse rise time.
The sharper pulse the bigger gap can be used and therefor to higher voltage capacitor can be charged.
Ded Ivan says that for air optimal is 50-100ns but I have achieved so far only 150-200ns and maximum COP about 50%.
The last picture from document was posted by some replicator from Ukraine. He spent several years to achieve 100ns rise time and he got COP 400%.
Of course, I understand that this could be just a false claims...but even if it is so, I don't see why this could not work as described
Best Regards