Optical Tolerancing and Sensitivity Analysis Explained
A nominal optical design assumes every surface, spacing, and material is perfect. Real parts are not. Optical tolerancing is the engineering that decides how much imperfection each element can have while the assembled system still meets its requirements, and it is often what separates a design that can be built at cost from one that cannot. It is a core part of any serious custom optical design.
Why tolerancing decides manufacturability
Every manufacturing process has variation: radius of curvature, thickness, wedge, surface figure, index of refraction, and the position of each element in its mount all vary from part to part. If the design is too sensitive to that variation, few assembled units will meet spec, yield collapses, and unit cost rises. Tolerancing quantifies that relationship before any parts are cut.
Sensitivity analysis
Sensitivity analysis asks how much the system performance, such as modulation transfer function, wavefront error, or boresight, changes when each parameter is perturbed by a small amount. It ranks the parameters from most to least critical.
This ranking is where the value is. A handful of parameters usually dominate performance, and the rest barely matter. Tightening the critical few and loosening the rest is how a designer buys performance where it counts and saves cost everywhere else. Tightening everything uniformly is the classic way to make a design needlessly expensive.
Tolerances and compensators
Tolerances are the allowed range for each parameter. They are assigned so that realistic manufacturing processes can hold them, informed by the sensitivity ranking and by what suppliers can actually achieve.
Compensators are adjustments, such as a focus movement or a chosen spacer, used during assembly to recover performance lost to part variation. Deciding which adjustments to allow is part of the tolerancing strategy, because a good compensator can dramatically loosen otherwise tight tolerances.
Monte Carlo yield
Monte Carlo analysis builds thousands of virtual assemblies, each with parameters drawn randomly from the tolerance ranges, and predicts the distribution of performance. The result is a yield estimate: what fraction of built units will meet spec.
The value of Monte Carlo depends entirely on realistic input distributions. Optimistic assumptions produce precise-looking yield numbers that collapse at first article. The distributions should reflect credible supplier processes, not best-case guesses.
As-built performance and design transfer
The output of tolerancing is an as-built performance prediction and a set of drawings and specifications a supplier can build to. This is the bridge between design and manufacturing, and it feeds directly into optical design transfer to manufacturing.
Done well, tolerancing is not paperwork. It is how a designer proves, before committing to hardware, that the system can be built at the required performance, yield, and cost. PAO tolerances designs in Zemax and CODE V and can develop the tolerance and yield strategy alongside the optical design.