"50:50" is the most quoted and least meaningful number in the beamsplitter catalogue. It describes a measurement, not a component — and the measurement has at least five variables attached to it.

Variable one: wavelength
A dielectric beamsplitter splits by interference, so its ratio is a function of wavelength. A plate specified at 50:50 for 550 nm will not be 50:50 at 532 nm, and it will be nowhere near 50:50 at 1064 nm. Broadband beamsplitters exist, but they achieve flatness by flattening the whole curve, which usually costs extinction and increases layer count.
Variable two: angle
Ratio moves with angle of incidence, and it moves faster than most drawings assume. A 45° plate used at 43° or 47° — which happens as soon as it is mounted in a real system with real tolerances — will shift by a percent or two. For a ratiometric measurement, a percent or two is a systematic error.
Variable three: polarisation
At non-normal incidence, s- and p-polarised light see different admittances and therefore different ratios. A "50:50" plate at 45° might be 45:55 for s and 55:45 for p. If the source is unpolarised and the detector is polarisation-sensitive, the two effects compound. This is the origin of the number that appears in better datasheets: polarisation splitting, or the difference between the s and p ratios at the design angle.
Variable four: position on the part
Coating thickness varies across a chamber. On a 25 mm plate the variation is small; on a 100 mm plate it can be enough to move the ratio by a percent or more from centre to edge. For an imaging system that uses the full aperture, the ratio is not a single number — it is a map.
Variable five: the measurement itself
Ratio is measured with a spectrophotometer at a defined spot, angle and polarisation, with a stated uncertainty. Two suppliers quoting 50:50 ± 3% may be measuring under different conditions, and the numbers are not comparable. Ask for the measurement geometry.
Cubes versus plates
A cube beamsplitter eliminates the second-surface reflection and the astigmatism that a plate introduces in a converging beam, at the cost of a cemented interface, thermal expansion mismatch and a higher price per part. A plate is cheaper, lighter and easier to mount, but it displaces the beam and produces a ghost unless it is wedged or AR-coated on the second surface.
The choice is usually dictated by whether the beam is collimated. Collimated: plate, with a wedge. Converging: cube, or a plate thin enough that the astigmatism is inside budget.
Non-polarizing is a design, not a promise
A "non-polarizing" cube achieves its flat ratio by combining a coating design with a cemented interface and, in some designs, a phase-compensating element. It works over a stated band and angle range. Outside them, it polarises. Buyers who need a genuinely polarisation-independent split over a wide band should expect a more complex — and more expensive — component, and should ask for the s and p curves separately rather than a single ratio.
What to put on the drawing
- Ratio, at a named wavelength, angle and polarisation state.
- Ratio tolerance, and whether it applies at a point or across the aperture.
- Polarisation splitting limit, if the application is polarisation-sensitive.
- Angle of incidence range, and whether the beam is converging.
- Transmitted wavefront and surface flatness after coating.
- Ghost and second-surface reflection requirements.
A beamsplitter is a two-output component. Specifying one output and ignoring the other is how systems end up with a 50:50 split and a 10% energy budget surprise.