What aspects of the design of a beamsplitter cube determine wavefront error/distortion? I was recently looking into having some custom beamsplitter cubes made. I require good wavefront quality, so I wanted to ensure that this was a primary consideration of the design (see how Thorlabs has listed "wavefront error" in their beamsplitter cubes). But what aspects of the design of a beamsplitter cube determine wavefront error/distortion (that is, how do I minimize wavefront error/distortion)? I would like to learn this for my future design efforts.
 A: Ultimately it's going to be application specific.
Specifications for manufacturing using real-world materials and processes is where the problems arise. Entrance and exit and diagonal interface surface flatnesses will probably be the dominant problem assuming the prisms used to make the cube are of high quality optical material. Specifying flatness is not a trivial task since you can have deviations at all spatial frequencies.
For example a sinusoidal $\lambda / 10$ deviation with a period of a few dozen wavelengths makes a diffraction grating which might technically preserve wavefront flatness of each diffracted beam but wreak havoc for your application.
So your flatness specification will need both an amplitude and a spatial frequency component; in fact it may be a plot of amplitude vs spatial frequency, a bit like how vibration specifications are done for optical tables.
You might have $\lambda / 10$ at 10 lines/cm but only 1 nm at 1000 lines/cm for example.


Figure 2: A Typical Compliance Curve of a High-Performance Thorlabs Optical Table

Source
