For a decade the industry trend was toward air-spaced assemblies. Cemented doublets were replaced by mounted elements, polarising cubes by wire-grid plates, and the adhesive layer was treated as a legacy detail. That trend has reversed, quietly and for practical reasons. Compact projection engines, waveguide combiners, fluorescence cubes and machine-vision splitters all need two surfaces held together at an angle with a controlled optical interface, and air spacing cannot deliver that in the space available.

The consequence is that a process most optical engineers last thought about in university is now a yield limiter, and it is being specified by people who have never had to qualify an adhesive.
The adhesive is an optical design decision
A cement layer is a thin optical element with its own refractive index, dispersion and thermal behaviour. Choosing it after the glass is chosen is backwards. For a B270 cube at n equals 1.523, a UV-curable adhesive at 1.56 sits close enough that the interface reflection is negligible; a silicone at 1.41 does not, and the interface becomes a partial reflector whether the design wanted one or not.
Index matching is also wavelength dependent. An adhesive matched at 550 nm may be 0.01 away from the glass at 850 nm, which produces a reflection that a visible-range design never noticed and an NIR design cannot tolerate. For assemblies that must work from 400 to 1,000 nm, the adhesive's dispersion curve belongs in the tolerance stack.
Cure shrinkage and transmitted wavefront
UV-curable adhesives shrink two to five percent by volume as they polymerise. In a 3 mm thick layer that is a fraction of a micrometre, which sounds negligible until it is applied to a 20 mm aperture with a lambda/4 transmitted wavefront requirement. The shrinkage is not uniform either: it follows the illumination profile of the curing lamp, so a cube cured from one side develops a slight wedge.
The fix is process rather than material. Slow, multi-angle curing at lower intensity reduces the gradient; a post-cure thermal soak stabilises the final shrinkage; and the wavefront is measured after curing, not before. Suppliers who quote transmitted wavefront on an uncemented pair are quoting a number that will not survive the process.
Stress birefringence in polarising cubes
A polarising beam-splitter cube adds a requirement that ordinary cemented assemblies do not have: the adhesive must not become birefringent under cure stress. A few nanometres per centimetre of retardance in the cement layer is enough to leak p-polarised light into the reflected channel, which in a projection engine shows up as a contrast loss that cannot be corrected downstream. Low-stress adhesives exist, but they cure slowly, and slow curing is expensive. That trade-off is the reason PBS cubes are among the least commoditised parts in the catalogue.
The interface that fails first
Field failures rarely occur in the bulk of the adhesive. They occur at the glass-adhesive boundary, and they are almost always driven by moisture or by thermal expansion mismatch. Damp heat at 85 degrees C and 85 percent relative humidity for 1,000 hours is the standard screen; it attacks the bond line from the edge inward, and the failure looks like a milky fringe that grows over months.
Thermal cycling is the second mechanism. A projector that runs from ambient to 60 degrees C every day for five years accumulates thousands of cycles, and each cycle loads the bond line in shear. Adhesives with a glass transition temperature near the operating range are the ones that fail, because the modulus changes right where the stress is highest.
Laser exposure is the third, and the least discussed. Many UV-cure adhesives yellow under sustained 405 or 450 nm illumination at high power density, and some darken under pulsed 1064 nm. For any assembly in a laser path, the adhesive's damage threshold belongs in the specification alongside the coating's.
Testing that predicts field life
Four tests, run on production parts rather than coupons: damp heat per IEC 60068-2-78, thermal shock from minus 40 to plus 85 degrees C, a power cycling test at the actual irradiance, and a shear strength check on a sacrificial sample from each lot. Add a visual inspection under collimated light for bubbles above 50 µm and for edge voids, because both are stress concentrators and both are invisible in transmission under diffuse light.
Buying cementing as a capability
The commercial question follows from all of this. An optics supplier that polishes and coats but subcontracts cementing has split responsibility for the final wavefront, and when a cube fails the interface test, the two suppliers will point at each other. A supplier that cements in-house, cures under process control and measures transmitted wavefront on the finished part owns the result.
That is the argument for treating cementing as a capability to be qualified rather than a step to be purchased. It is also why the drawings that specify adhesive grade, cure schedule and post-cure inspection get better parts than the ones that specify only glass type and dimension.
Product page: cemented beam-splitter assemblies — specifications, coating options and lead times.