Glossary#
This page defines the core concepts in Optiland. Want runnable examples? See the API Cheat Sheet.
Optic#
The central container for an entire optical system. An Optic instance holds all surfaces, aperture
definitions, field points, and wavelength information. It also exposes paraxial analysis, aberration
computations, the ray tracer, and polarization state.
Example: lens = optic.Optic(name="My System")
SurfaceGroup#
Manages the ordered collection of Surface objects within an Optic. It propagates rays through
the system by invoking surface-specific logic at each step, aggregates ray-trace history, and exposes
methods for adding, removing, and modifying surfaces.
Surface#
Represents a single optical interface — a lens element, mirror, image plane, grating, etc. Each surface is composed of:
Geometry: The mathematical shape (e.g., planar, spherical, aspheric, freeform).
Material: Refractive index and extinction coefficient on each side.
Coating: Optional thin-film stack modifying reflection, transmission, or polarization.
Interaction model: How rays interact — refraction, reflection, diffraction, or a custom phase profile.
Physical aperture: Optional mask defining the clear aperture.
BSDF: Bidirectional scattering distribution function for surface scatter.
Special surface types include ObjectSurface (first surface, object plane), ImageSurface
(final surface, image plane), and surfaces carrying a ThinLensInteractionModel (paraxial thin-lens
approximation).
Material#
Defines the optical medium between surfaces. Optiland supports:
MaterialFile: Loads dispersion data from the refractiveindex.info database, which is bundled with the package.IdealMaterial: A wavelength-independent medium specified by a single refractive index.AbbeMaterial: A medium specified by its refractive index at the d-line (nd) and Abbe number (Vd), using a Buchdahl dispersion model (recommended) or a legacy polynomial model.User-registered materials via
MaterialRegistry.
Geometry#
Defines the mathematical shape of a surface and provides two critical operations: ray–surface intersection and surface normal computation. Built-in geometries include:
StandardGeometry: Spherical and conic surfaces.EvenAsphere,OddAsphere: Polynomial aspheres.Biconic,Toroidal,Polynomial,ChebyshevPolynomialGeometry,ZernikePolynomialGeometry: Freeforms.PlaneGrating,StandardGrating: Diffraction gratings.NURBSGeometry: Non-Uniform Rational B-Splines.ForbesGeometry: Q-polynomial freeform surfaces (Forbes convention).
Custom geometries can be added by subclassing BaseGeometry (closed-form) or
NewtonRaphsonGeometry (iterative).
Aperture#
Defines the system’s limiting aperture. The aperture type determines how the entrance-pupil size is specified:
EPD: Entrance pupil diameter in mm.imageFNO: Image-space f-number.objectNA: Object-space numerical aperture.float_by_stop_size: The aperture stop physical diameter drives the pupil size.
Fields#
Define the points in the object plane (or angular directions) being imaged. Fields can be specified by
angle (degrees) or object height (mm). Vignetting factors can be applied per field. Each Field has an
optional weight used in weighted analysis aggregation.
Wavelengths#
Specify the wavelengths of light used for analysis. All values are stored internally in microns (µm).
One wavelength is designated as the primary wavelength, used for paraxial calculations and single-wavelength
analyses. Each Wavelength has an optional weight used in weighted analysis aggregation.
Coordinate System#
Each surface has its own Local Coordinate System (LCS) defined by position (x, y, z) and rotation (rx, ry, rz) relative to a reference. Key conventions:
Light propagates from left to right along the +z axis.
Surface 1 is typically at the global origin (z = 0).
Thickness is the axial separation to the next surface; positive means to the right.
Radius of curvature: positive means center of curvature to the right (convex to the incoming beam); negative means to the left.
Tilts and decenters are applied as
R = Rz @ Ry @ Rx.
Apodization#
Defines the intensity (amplitude) distribution across the entrance pupil. The default is
UniformApodization (flat). GaussianApodization applies a Gaussian intensity profile across the
pupil, modelling a Gaussian input beam.
Backend#
Optiland routes all numerical operations through a unified backend abstraction (optiland.backend),
allowing transparent switching between NumPy (default, CPU) and PyTorch (GPU and autograd). All
Optiland code uses import optiland.backend as be instead of importing NumPy or PyTorch directly.
Switch backends with be.set_backend("torch") or be.set_backend("numpy").
See Configurable Backend in the Developer’s Guide for the full backend architecture.
Non-sequential ray tracing#
A tracing mode where rays propagate freely through a 3-D scene and interact with surfaces in any order (rather than a fixed numbered sequence). Used for illumination, stray-light, ghost, and non-imaging analysis. See the NSQ gallery.
Monte Carlo tracing#
Sampling many random rays from sources and accumulating their contributions statistically. NSQ uses Monte Carlo sampling; results converge as the ray count increases.
Splatting#
How a detector deposits a ray’s flux into its pixel grid. hard splatting bins to the nearest
pixel; bilinear splatting distributes flux across the four nearest pixels and is differentiable
with respect to the landing position.
Throughput (weight)#
A per-ray multiplicative factor on flux that accumulates as a ray reflects, refracts, or scatters. NSQ’s differentiable Fresnel split uses an attached throughput weight whose forward value is 1 but which carries gradients (see detached-sample / attached-weight).
BSDF#
Bidirectional Scattering Distribution Function — the model that determines how a surface scatters incident light into outgoing directions (e.g. specular, Lambertian, Harvey–Shack roughness, or tabulated measured data).
Irradiance map#
A 2-D spatial distribution of flux per unit area [W/mm²] recorded on a detector surface.
Far-field pattern#
The angular distribution of radiant intensity [W/sr] emitted into a hemisphere, recorded by a
FarFieldDetector.
Étendue#
A conserved geometric quantity (area × projected solid angle) describing how “spread out” a bundle of light is in space and angle. It bounds what any non-imaging concentrator can achieve.
Detached-sample / attached-weight#
The estimator NSQ uses to keep stochastic reflect/refract choices differentiable: the branch is sampled from a detached (non-differentiable) probability, while the throughput weight stays attached to the autograd graph so gradients flow through material and geometry parameters. See the NSQ developer guide.
Visibility gradient#
The gradient contributed by a moving silhouette or occlusion boundary (e.g. vignetting at an aperture edge). NSQ v1 does not capture these (they are zero); see NSQ Limitations & Roadmap.