Polarizing beamsplitter
Separates orthogonal polarization states into two paths.
Open in the canvas →In the real world
An ordinary beamsplitter divides a beam by intensity and does not care how it is polarized. A polarizing beamsplitter divides it by polarization instead: one linear state is transmitted, the orthogonal state is reflected, and — unlike a polarizer, which absorbs or dumps what it rejects — both halves leave as usable beams. Nothing is thrown away, which is what makes the device a router rather than a filter.
The usual form is a cube: two right-angle prisms cemented along their hypotenuses, with a multilayer dielectric coating sandwiched between them. Light meets that internal interface at 45°, and the layer stack is designed so that the p-polarized component (electric field in the plane of incidence) is transmitted while the s-polarized component is reflected through 90°. The two outputs are therefore linearly polarized and perpendicular to one another.
How the incoming power divides follows Malus's law, so the split is set by the input polarization angle rather than by the cube:
Putting a half-wave plate in front turns this into a continuously variable beamsplitter: rotating the plate rotates the input polarization, sweeping the split from all-transmitted to all-reflected without any absorption anywhere. That pairing is one of the most common two-element combinations on an optical bench, used for power control, for balanced splitting, and for routing a beam between two experiments.
Run backwards, the same cube combines two orthogonally polarized beams into one path — the standard way to overlap two lasers with no loss, which no intensity beamsplitter can do.
One asymmetry matters in practice. The transmitted port is usually very pure, with extinction ratios of 1000:1 or better, because the coating is good at rejecting s. The reflected port is markedly worse, often nearer 20:1, since some p light leaks into it. If an experiment needs a clean state, take it from the transmitted port, or clean the reflected one up with a polarizer afterwards.
In OpticalSetup
The cube is drawn with its coated diagonal, and that diagonal is the traced surface. It transmits horizontal polarization along the incoming axis and reflects vertical polarization through 90°, splitting a single incoming ray into two outgoing beams that the tracer follows independently.
The division follows Malus's law exactly: linear light at 0° goes fully through, at 90° fully across, at 45° splits half and half, and at 30° divides 75/25. Unpolarized light splits evenly, as it should. Both outputs emerge in pure linear states — horizontal on the transmitted port, vertical on the reflected one — regardless of what arrived, which is what makes a PBS a polarization cleanup element and not merely a splitter. A port receiving less than 2% of the light is dropped rather than drawn as a hairline that suggests a beam nobody could use.
Because the cube resolves polarization into two paths, it is also how the sketch makes polarization visible: put one after a half-wave plate and rotating the plate's axis visibly shifts power from one output arm to the other, with no attenuation anywhere in the path.
The cube also handles fast polarization switching properly. A beam alternating between two states pulse by pulse leaves each port as a genuinely gated pulse train, with the two ports complementary — so an electro-optic modulator followed by a PBS produces two real interleaved trains rather than two steady half-power beams.
The cube is ideal. Both ports are perfectly pure, which the reflected port of a real cube is emphatically not — expect nearer 20:1 there — so an experiment whose result depends on reflected-port purity will look better here than on a bench. There is no coating loss, no residual reflection at the entrance and exit faces, and no angular or spectral acceptance: the split is the same at every wavelength and every angle of incidence, whereas a real cube is specified for a band and degrades outside it. The glass path through the cube is not modelled either, so it adds no optical path and no group-delay dispersion to a pulse.