Both are factors in good centering, but mostly the change in diameter. In turns, there is a natural tendency for the train to shift towards the outside of the curve due to inertia. The wheel diameters become asymmetric which helps to re-center the train. It's usually not sufficient on its own, which is why superelevation is used as well - the outside rail is somewhat higher than the inside rail which shifts relative gravity to pull the train back towards the inside as well. The relationship between these two effects is a bit complex (depends on weights and speeds of trains) so it's usually all a bit approximate.
The conical section of the wheels is mostly intended to prevent hunting on straight track, and the shape can't be made too aggressive without increasing the wear on wheels on rails. So on curves the superelevation is added to provide the extra force required.
Because conical wheels do increase wear and can contribute to oscillation in their own way, there have been experiments with cylindrical wheels especially on higher-speed trains---BART is a well known example. It ultimately didn't work very well and so they have been re-trueing the wheels to a non-cylindrical profile, although still not quite a traditional conical one. Basically in higher-speed operation the re-centering effect is too significant and causes one wheel to "chatter," which over time creates a significant vibration in the rail. Trouble is cylindrical wheels tend to cause the same thing to happen on the other side. It was a very hard problem before computer modeling became available.
> It's usually not sufficient on its own, which is why superelevation is used as well - the outside rail is somewhat higher than the inside rail which shifts relative gravity to pull the train back towards the inside as well. [...] So on curves the superelevation is added to provide the extra force required.
I've never heard about that theory as for why superelevation/cant is supposedly being used until now.
Given that most of the time you'll end up with a remaining net force to the outside of the curve even after application of cant, it doesn't seem to make that much sense, either.
That's the conventional explanation of superelevation, although I worded it in sort of an odd way. But I'm describing the same thing that e.g. Wikipedia does. Superelevation directs the force of the car more "straight down" in relation to the rails which improves centering and balance of the load by the same token. The thing I said about "shifting gravity" is unnecessarily confusing because it depends on reference frame.
I think for low-speed freight the balance needs to be pretty close on to ideal to meet regulations, e.g. FRA regulations give calculations for acceptable ranges. But since it's dependent on running speed it's hard to get correct for freight and passenger mixed operation which is the subject of this FRA report that has a lot of detail on the calculations: https://railroads.dot.gov/sites/fra.dot.gov/files/fra_net/19...
Hmm, well the conventional explanation that I know of is that it's simply to reduce the lateral forces acting on your train and more importantly on the payload you're carrying - especially with passenger trains it's passenger comfort that's the limiting factor by far, not safety against derailment or overturning (which is how tilting trains can work, since tilting the train body only reduces the forces felt inside the passenger compartment, but not the forces acting at the wheel-rail level).
I see what you mean with regards to how it's also described on Wikipedia – only I've got some currentish (European) literature in front of me which claims that cant and the resulting cant deficiency/excess are only of secondary importance with regards to wheel and rail wear (the main factors are simply the curve radius itself and the construction of the running gear of the trains operating over the curve), and as such the main importance of cant is simply ride comfort. Likewise it also claims that according to some practical experiments done by some infrastructure operators, no link could be found between occurrences of cant excess for slower moving heavy freight trains and increased maintenance requirements (Which interestingly somewhat contradicts the corresponding supposition given in your FRA document...).
This also matches the evolution of the design rules on the German national railways – in the 80s there still used to be a relatively elaborate system of determining the allowable cant excess for slower moving trains depending on the annual tonnage of that kinds of trains, but since then at some point that system got dropped and has been radically simplified:
The regular cant is simply 55 % of the equilibrium cant and it's up to the design engineer to deviate from that value if necessary (when the speed distribution varies from that of a normal mixed-traffic route).
Interestingly all of that somewhat contradicts the statements given in your linked FRA document. To some extent this can probably be explained by European freight trains being shorter, somewhat lighter (lower axle loads) and also nowadays slightly faster than their American counterparts, and also due to traditionally using somewhat higher allowable cant deficiency values, especially with regards to passenger rolling stock.
It likely doesn't explain everything, though, but I don't know enough, either, to reconcile those two differing points of view.
The conical section of the wheels is mostly intended to prevent hunting on straight track, and the shape can't be made too aggressive without increasing the wear on wheels on rails. So on curves the superelevation is added to provide the extra force required.
Because conical wheels do increase wear and can contribute to oscillation in their own way, there have been experiments with cylindrical wheels especially on higher-speed trains---BART is a well known example. It ultimately didn't work very well and so they have been re-trueing the wheels to a non-cylindrical profile, although still not quite a traditional conical one. Basically in higher-speed operation the re-centering effect is too significant and causes one wheel to "chatter," which over time creates a significant vibration in the rail. Trouble is cylindrical wheels tend to cause the same thing to happen on the other side. It was a very hard problem before computer modeling became available.