Considering the Taylor series with remainder,
It is known that the function evaluations f(x+h) and f(x-h) in floating arithmetic are not exact.
Assuming that Floating Point evaluations are the respective floating point evaluations of f(x+h) and f(x-h), where ϵ< ϵ_MACH.
How to go about computing the maximum roundoff error?
I have tried working this out and I have been coming to the conclusion that the maximum error is ϵ_MACH. However, I doubt that this is correct.
Would anyone mind showing me how to work this out?