Do filters break when cutoff is at nyquist ?

[Marvinh]

I have had some instances where it’s completely fine, and other times have had inf or nan appear . So I’m not entirely sure if oversampling is required .

I usually work myself at 44100 and have seen cutoff values at 22050 in other plugins .

[kerfuffle]

It depends on the filter and how the coefficients are calculated. But even if the plug-in says the cutoff is at 22050 that doesn't mean it internally goes that high, they might be clamping the cutoff to some value just below Nyquist.

[mystran]

I have had some instances where it’s completely fine, and other times have had inf or nan appear . So I’m not entirely sure if oversampling is required .

I usually work myself at 44100 and have seen cutoff values at 22050 in other plugins .

You can't put poles above Nyquist. If you try you'll typically end up with infinities or NaNs (either in the coefficient computation, or due to an unstable filter).

You can attempt to fit a response to a filter that would have it's poles at a higher frequencies (tends to be most useful for EQs and such) or you can oversample, which pushes the Nyquist higher so we're no longer above it. You can also cheat and not compute your coefficient correctly, in which case your actual cutoff is lower than it should be; this might sound like it's an objectively bad thing to do, but in some cases it can actually work out quite well (especially when also combined with oversampling).