"Pointwise L2 error"

@beams2

Carried forward from !16 (merged).

There's one wrinkle in here (that comes from your original example): IMO, taking relative errors pointwise and then computing an L2 norm of them doesn't make a whole lot of sense--what if the true solution passes through zero? Instead, do \|error\|_2/\|true\|_2. Could you fix that?