There are some open questions, but I'll take a stab at answering. I'll assume you want voltage control of a load that must be ground referenced, the voltage range is 0-45V, a 48V supply is available, the maximum current is 1A, and the control input is a voltage from 0-5V.
Here is a circuit that fullfills the requirements I stated:
This is similar in idea to Russell's circuit with a few key differences. Q2 is a controlled current sink linear with the opamp output voltage in the range of about 600mV to 5V. This current variably turns on PNP transistor Q1. The opamp output from about 600mV to 5V maps linearly to the load current, which should help stability. The compensation cap C2 working against R2//R5 provides a means to add additional stability as needed. C2 shouldn't need to be more than a few 10s of pF.
With 5V on the base of Q2, the emitter will be about 4.3V, so Q2 will sink 70 mA. Assuming the power transistor Q1 has a gain of at least 15 (in the plausible range for this type of transistor), the load current can be up to 1A.
R2 and R5 divide the load voltage into the 0-5V range the opamp can handle. Since stuff happens, you want to make sure all is OK with the full 48V at P1. This 48V divided by R2 and R5 becomes 4.75V into the opamp. That's close enough to 5V to use most of the range but still leave a little margin.
You will have to think about the power dissipation of Q1 carefully. It could be quite a lot depending on what current your load really draws. Worst case the load voltage is half the supply, so 24V, and drawing 1A. That puts 24W on Q1, which is quite a lot. If your load really can draw up to 1A, then Q1 probably should be a TO-3 with a good heat sink and forced air cooling. If that's too much, you need to consider switching topologies to accomplish what you are doing. 24W is not trivial to deal with.
Q2 could also get toasty, but nowhere near as bad as Q1. At the maximum of 5V on it's base, it will drop about 43V at 70mA, which is 3W. That's not too hard to deal with, like a TO-220 with a small heat sink. Of course if your load doesn't really need 1A this all scales down linearly.
Oops:
I updated the schematic to get rid extra resistor in series with the opamp negative input. The circuit evolved as I was drawing it and I didn't notice this resistor was no longer needed when the circuit was posted originally. The description has been updated accordingly.
One way is to use a high-voltage probe. For example if you check the B&K PR 28A Datasheet you'll see it's rated for 1kV to 40kV DC and 1kV to 28kV RMS AC. As a ballpark price they are currently $US72 at Digikey. They include a 1000:1 divider to bring the voltage down to levels that normal test equipment can deal with.
I have seen DIY solutions in the past, but of course you'd have to be very careful with the materials used and the construction so I wouldn't really recommend it. Especially if you don't have to knowledge / equipment to test insulation breakdown.
Best Answer
15kV is a very large voltage. Given any current path it will arc over the air. You may need to use conformal coat to prevent moisture on the surface of components causing this to happen. Needless to say 15kV is very dangerous. Place high value blead resistoes over any part of the circuit that is capacitive. Even a few pF at 15kV will give you a nasty shock.
You can use cascode, but the biasing will be a real headache. The bias tree has to be at many kV end to end. Any leakage here will wipe out your PSU which I expect can't deliver much continuous current.
Transformer gate drive isolation is better (though you need a really good primary/secondary insulation) But once you have that the gates of each FET are only at a few Volt above their source.
You stack FETs, Source to Drain e.g. 20 fets each 750V. Buffer and bias each FET gate with another small FET and a rectifier (the extra FET gives better turn off times by draining gate charge). This rectifier receives current from a small transformer.
This transformer's primary is a large very insulated wire driven with a large AC current, e.g. 10-15A
You can even do push-pull this way using two trees of FETs. (No need for P channel). Drive each side with an anti-phase signal (wind transformers in anti-phase) with electronics ensuring a turn off gap to prevent shoot through.
This will easily switch 15kV at a reasonable current at a few kHz.
This document gives some of the info you require. Their example goes to 5kV.
I once worked on a system where we had several stages, each switching 1kV and used isolated DCDC converters to pass power between stages (so each no more than 1kV from the on next to it) and an optical fibre passing the trigger signal which magnetically coupled to the tree of FETs. This switched 60kV at 10's of kHz and hundreds of amps!!!