en.wikipedia.org/wiki/Fourier-transform_infrared_spectroscopy

This statement presents a power-of-two length as a requirement, but modern FFT implementations do not require that.

The core mathematical task here is a discrete Fourier transform (DFT). Well-established FFT libraries compute DFTs for arbitrary lengths, including prime sizes. For example, the official FFTW documentation states that "the input data can have arbitrary length" and that FFTW uses O(n log n) algorithms "for all lengths, including prime numbers." Its reference manual also says of the transform size n: "It can be any positive integer," while noting only that powers of 2 are especially fast.

So the accurate statement is that power-of-two lengths were historically convenient for some radix-2 FFT implementations and may still be performance-optimal in many cases, but they are not required for rapid FFT-based spectral calculation in general.

This sentence misstates the unequal glass paths in a standard Michelson interferometer that uses a plate beamsplitter.

Authoritative optics references explain that the compensator plate is needed because the two arms do not traverse the beamsplitter substrate equally: one beam goes through the beamsplitter glass three times, while the other goes through it once. OpenStax's University Physics text says, "one beam passes through M three times and the other only once," and notes that the compensator plate is added so both beams traverse the same thickness of glass. TYDEX's FTIR beamsplitter documentation gives the same explanation, stating that the compensator plate corrects for the fact that "the other beam passes through the beamsplitter plate three times instead of one."

Because both cited sources describe the unequal traversal as three versus one, the article's two versus one wording is factually incorrect.