Geometrical optics is the branch of optics that describes the propagation of light in terms of rays: directed curves representing local propagation paths. It explains reflection, refraction, image formation, and the transport of optical power without explicitly resolving the oscillating electromagnetic field. Its central approximation is that the wavelength is small compared with the dimensions of optical components and the distances over which the properties of a medium change appreciably.
The subject provides the working language of lenses, mirrors, prisms, cameras, telescopes, microscopes, and illumination systems. It also describes continuously curved paths in the atmosphere and graded-index materials. Although rays do not by themselves explain diffraction or interference, geometrical optics is not a competing theory of light: it is a short-wavelength limit of wave theory. Its practical strength lies in separating the geometry of propagation from the more detailed calculation of phase, polarization, and wave effects.
Rays, optical paths, and the short-wavelength approximation
A ray specifies a position and a direction of propagation. In a homogeneous, isotropic medium, rays are straight lines; in an inhomogeneous medium, they generally curve. A collection of nearby rays forms a ray bundle, whose changing cross-sectional area helps determine how optical power is distributed. A mathematical ray has no width, so an individual ray is not a physically isolated, infinitely narrow beam.
The material quantity governing propagation is the refractive index, , where is the speed of light in vacuum and is the phase velocity in the medium. The optical path length along a curve is
where is an element of geometrical distance. In a uniform material, a physical length therefore contributes an optical path length . Optical path measures accumulated phase: for vacuum wavelength , the phase accumulated along a path is , apart from phase changes introduced at boundaries.
Fermat’s principle states that a physical ray makes the optical path stationary with respect to small variations of its route, with appropriate endpoint and boundary conditions. “Stationary” is more general than “shortest”: the first variation vanishes, but the path need not be the unique global minimum. For a monochromatic field in a stationary medium, this formulation concerns phase optical path; in dispersive materials, the travel time of a pulse involves the group velocity and requires additional care.
A wave-based derivation begins by representing a monochromatic scalar field as a slowly varying amplitude multiplied by a rapidly varying phase,
Substitution into the wave equation and retention of the leading terms at large gives the eikonal equation,
Surfaces of constant are wavefronts. In an isotropic medium, rays are normal to these surfaces. The next order of approximation determines amplitude transport along the ray bundle.
This construction explains both the power and the limitations of ray descriptions. Straight-line propagation is a consequence of constant refractive index, rather than a universal property of light. Likewise, geometrical optics becomes unreliable where the amplitude or material properties vary on wavelength scales, or where a ray construction predicts singular behavior. In anisotropic media, the direction of energy transport need not coincide with the wavefront normal, so wave-normal and energy-ray descriptions must be distinguished.
Reflection, refraction, and optical boundaries
At a smooth interface, reflection sends an incident ray back into its original medium. The incident ray, reflected ray, and surface normal lie in one plane, and the angles measured from the normal satisfy
A plane mirror consequently forms an image that appears as far behind the mirror as the object lies in front of it. Curved mirrors redirect different incident rays according to their local surface normals, allowing a surface to focus or spread a bundle.
For transmission between isotropic media, refraction obeys Snell’s law,
A ray entering a higher-index medium bends toward the normal; one entering a lower-index medium bends away from it. This law follows either from stationarity of optical path or from continuity of phase along the interface. The latter derivation expresses conservation of the tangential component of the wave vector. At normal incidence, a change of refractive index changes the wavelength and phase speed but not the ray direction.
When , a transmitted propagating ray exists only up to the critical angle,
Above this angle, total internal reflection occurs. For glass of refractive index approximately 1.50 adjoining air, the critical angle is about . This effect is used in prisms, light guides, and optical fibers. The wave description nevertheless contains an evanescent field beyond the interface, so total internal reflection does not imply that the electromagnetic field vanishes immediately outside the denser medium.
Ray directions alone do not specify how incident power divides between reflected and transmitted branches. The Fresnel equations, obtained from electromagnetic boundary conditions, supply the amplitude coefficients and their dependence on incidence angle and polarization. For a nonabsorbing interface at normal incidence, the power reflectance is
An uncoated air–glass surface with therefore reflects approximately 4 percent of normally incident power. Multiple such surfaces can produce appreciable losses and unwanted ghost images.
Surface quality also matters. Specular reflection assumes that the boundary is sufficiently smooth on the relevant optical scales. Roughness can redistribute power through scattering, while thin coatings alter reflection through interference. Practical ray models commonly combine geometrical directions with wavelength-dependent reflection and transmission coefficients, but calculating a coating’s behavior requires wave theory rather than Snell’s law alone.
Image formation by lenses and mirrors
An optical image is formed when rays from each object point converge toward, or appear to diverge from, a corresponding image point. A real image involves actual convergence and can be received on a screen or detector. A virtual image is located by extending outgoing rays backward; no light need pass through that apparent location. An ideal image-forming system would preserve this point-to-point correspondence throughout a specified object region.
A lens forms images by refraction at its surfaces. For a thin lens in air, using positive object distance for a real object and positive image distance for a real image, the Gaussian imaging equation is
where is focal length and and are object and image distances. The transverse magnification is
Thus, a converging lens of focal length , with an object away, forms a real image behind the lens, with magnification . The negative sign indicates inversion.
A thin-lens ray construction often uses three convenient rays: one initially parallel to the axis, one directed through the appropriate focal point, and one passing through the optical center with negligible change of direction. These are bookkeeping devices rather than privileged physical rays. Their familiar construction rules rely on the thin-lens and small-angle approximations.
The focal length depends on both material and surface curvature. For a thin lens in air,
with radii positive when their centers of curvature lie to the right, assuming propagation from left to right. A lens’s converging or diverging action depends on its refractive index relative to the surrounding medium. A glass lens placed in a sufficiently high-index liquid can therefore behave differently from the same lens in air.
For a spherical mirror, the paraxial focal length has magnitude , where is the radius of curvature. A paraboloidal mirror instead brings axis-parallel rays to a common geometrical focus without spherical aberration. Neither statement guarantees perfect imaging for arbitrary off-axis objects.
Thick lenses and multi-element assemblies require principal planes, focal points, and nodal points rather than a single physical lens plane. Object and image distances in the Gaussian equation are then measured from the corresponding principal planes. These planes can lie outside the glass itself. Focal length is consequently not generally equal to the distance from the last optical surface to the image plane, a distinction important in telephoto and wide-angle photographic designs.
Paraxial optics and matrix methods
The paraxial approximation describes rays close to an optical axis and making small angles with it. Replacing and by , and by 1, turns many ray equations into linear relations. This first-order theory determines focal lengths, magnifications, image locations, and pupil positions without yet accounting for the leading image errors of finite-aperture systems.
In a centered system, a meridional ray can be represented by its height and reduced angle . A first-order optical element acts through a two-by-two matrix,
Propagation through distance in a uniform medium of refractive index has matrix
while an ideal thin lens in air has matrix
The matrix for several successive elements is their ordered product, with the first encountered element acting on the ray vector first. This makes complicated arrangements manageable using linear algebra.
For conjugate object and image planes, the system matrix has : the final height then depends on initial height but not on the initial ray angle. The coefficient gives transverse magnification. A system with is afocal, meaning that an initially parallel bundle remains parallel, although its width and angular orientation may change. Telescopes and beam expanders are important examples.
Two thin lenses in air, separated by distance , have equivalent power
Separation therefore changes system power even when the individual lenses remain unchanged. This is one reason a multi-element assembly cannot generally be treated as several powers simply added together.
For these matrices in reduced-angle coordinates, . This property is associated with conservation of a paraxial optical invariant. For two rays in the same centered system,
remains constant through ideal propagation and refraction. The invariant links image size to angular spread and anticipates more general throughput constraints.
Matrix methods also extend to resonators. For a real, lossless paraxial round-trip matrix, the usual nondegenerate stability condition is . The geometrical condition identifies whether small ray deviations remain bounded, although the transverse field structure of a laser resonator requires a wave treatment.
Aberrations and departures from ideal imaging
First-order optics predicts exact point images only within its approximations. Real optical systems use finite apertures and fields, and their image errors are described by optical aberrations. These can be expressed as ray-intercept errors in an image plane or as departures of the emerging wavefront from an ideal reference sphere. An aberration-free geometrical image can still have finite size because of diffraction, so aberration and diffraction are distinct limitations.
The classical third-order monochromatic aberrations are spherical aberration, coma, astigmatism, field curvature, and distortion. Spherical aberration occurs when rays passing through different zones of an aperture have different axial intersections. A spherical converging surface typically does not focus marginal and near-axis rays at the same location. Stopping down the aperture reduces the contribution of marginal rays, but also reduces collected light and can increase the relative importance of diffraction.
Coma affects off-axis points and produces an asymmetric distribution of ray intercepts. Astigmatism gives different focal positions for two principal orientations of an off-axis bundle, commonly described by tangential and sagittal image surfaces. Field curvature places best focus on a curved surface rather than a plane. Distortion changes the mapping between object position and image position without necessarily blurring an individual point; barrel and pincushion distortion are familiar photographic examples. Correcting one error does not automatically correct the others.
Chromatic aberration arises because refractive index varies with wavelength, a property called dispersion. Longitudinal chromatic aberration shifts the axial focus with wavelength, while lateral chromatic aberration changes image scale. An achromatic doublet combines glasses of different dispersion so that two selected wavelengths share a focal location. For thin elements in contact, the first-order conditions can be written
where and are Abbe numbers, measures of inverse relative dispersion. Residual wavelength-dependent errors remain outside the selected correction conditions.
Designers balance aberrations by adjusting surface curvatures, element spacing, glass selection, and stop position, and by introducing aspherical or freeform surfaces. A paraboloid eliminates spherical aberration for an on-axis object at infinity but still has off-axis aberrations. Reflective systems avoid the ordinary chromatic power associated with refractive lenses, although their coatings can remain wavelength dependent.
No single image-quality number captures every use. A spot diagram displays geometrical ray intersections, whereas wavefront error supports diffraction calculations. Permissible distortion may be large in an instrument whose geometric mapping is calibrated, yet unacceptable in a measurement system that assumes uniform magnification. Aberration specifications therefore depend on field, aperture, wavelength range, detector sampling, and the quantity being measured.
Apertures, pupils, and conservation of optical throughput
An optical system transmits only part of the light available from an object. The aperture stop is the element that limits the cone of rays from an on-axis object point. Its image as viewed through the optics before it is the entrance pupil; its image through the following optics is the exit pupil. A pupil is therefore an optical image of a limiting aperture, not necessarily a physical hole at the location where it appears.
The chief ray from an object point passes through the center of the aperture stop, while marginal rays reach its edge. These definitions help distinguish angular field from aperture. A separate field stop limits the extent of the object region that is transmitted. Vignetting occurs when some rays from an off-axis bundle are intercepted by additional mechanical or optical boundaries, producing field-dependent loss.
In photographic optics, the f-number is
where is entrance-pupil diameter. For otherwise comparable distant-object imaging conditions, image irradiance scales approximately as . Changing from to therefore admits about four times as much light, before allowance for transmission losses or vignetting. At close conjugates, magnification and pupil geometry modify the relation between the marked f-number and the effective image-side cone.
For microscopy and other high-aperture systems, numerical aperture is more useful:
where is the half-angle of the accepted cone in the relevant medium. Immersion objectives use a liquid between specimen and objective to achieve numerical apertures greater than those possible in air. Numerical aperture describes angular acceptance geometrically and also enters the wave-theoretical expressions for resolution.
A general measure of throughput is étendue, expressed for a bundle crossing an area as
Here is an element of solid angle. In ideal lossless propagation through reciprocal optical systems, étendue is conserved. Absorption and clipping remove power, while scattering can increase the angular or spatial extent occupied by the surviving distribution.
The corresponding radiometric result is conservation of along a lossless ray, where is radiance. Passive optics cannot arbitrarily compress a large, wide-angle source into a small, narrow-angle output while preserving its power. Increasing concentration necessarily increases angular spread or discards part of the input. This constraint determines the coupling efficiency of lamps into fibers, the design of projection systems, and the concentration achievable in solar collectors; it is a consequence of optical phase-space conservation rather than imperfect manufacturing.
Curved rays in graded-index and anisotropic media
Straight segments connected at surfaces are sufficient only when refractive index is piecewise uniform. In an isotropic medium with smoothly varying index , the ray equation is
where is arc length. Writing the unit tangent as gives
Only the part of the index gradient perpendicular to the ray changes its direction. Rays consequently curve locally toward increasing refractive index.
Atmospheric refraction is an important natural example. In the Earth’s atmosphere, refractive index depends on pressure, temperature, composition, and wavelength. A vertical index gradient bends rays, shifting the apparent altitude of astronomical objects. Close to the horizon, long atmospheric paths and variable thermal structure make the displacement particularly sensitive to local conditions. Astronomical refraction is therefore not represented accurately in all circumstances by one fixed correction.
Mirages arise when these gradients produce ray paths that connect an object and observer in unusual ways. Above strongly heated ground, lower-density air near the surface can have a lower refractive index than cooler air above. Rays can curve upward before reaching the observer, creating an apparent inverted image below a distant object. Superior mirages occur under other stratifications, notably temperature inversions, and may place apparent images above their ordinary positions.
Engineered graded-index components use the same principle. Near the axis of a cylindrical medium, a profile of the approximate form
produces approximately sinusoidal paraxial ray trajectories. The characteristic spatial period is . Rod sections of selected lengths can relay images or transform diverging bundles into nearly collimated ones without conventional curved refracting surfaces. Graded-index fibers can also reduce intermodal transit-time differences compared with comparable step-index fibers by giving outer paths a higher local propagation speed.
Anisotropic crystals require a different extension. Their optical response depends on direction and polarization, so a scalar index field is insufficient. Wavefront normals follow the dispersion relation, while energy rays follow the energy-transport direction, commonly associated with group velocity in transparent media. In birefringent materials, one incident beam can produce two differently polarized transmitted beams, and an extraordinary ray can exhibit spatial walk-off from its wave normal.
These effects do not invalidate geometrical reasoning, but they change its governing geometry. Accurate tracing must specify the material’s direction-dependent electromagnetic properties and track the appropriate wave branch, rather than applying isotropic Snell’s law with one numerical refractive index.
Optical instruments and computational ray tracing
Geometrical optics links component properties to instrument functions. In a camera focused on a distant scene, the detector lies near the rear focal plane, and focal length controls image scale. For an ideal rectilinear mapping, an object direction at angle has image height approximately
For a sensor dimension , the corresponding full angular field is approximately . Distortion changes this mapping, while aperture and detector sampling influence how much detail is usefully recorded.
An astronomical telescope combines a large entrance aperture with an image-forming objective or mirror. In an afocal visual telescope, objective and eyepiece focal lengths determine angular magnification; its magnitude is approximately . Exit-pupil size determines how the transmitted bundle fits into the observer’s eye. In a microscope, the objective’s numerical aperture and working distance are central constraints, while the objective and subsequent optics establish image scale. Modern infinity-corrected microscope objectives produce a nominally collimated output for on-axis object points, which a tube lens converts into an image.
Computerized ray tracing extends these constructions to complex systems. A tracing program determines a ray’s intersection with each surface, evaluates the local normal, and applies reflection or refraction. It may also propagate wavelength, optical path, polarization state, and a statistical power weight. Spherical surfaces permit relatively simple intersection calculations; aspherical and freeform surfaces frequently require iterative numerical solutions.
Sequential tracing assumes that rays encounter a prescribed series of surfaces. It is especially useful for lenses and other imaging assemblies. Nonsequential tracing instead determines which object a ray meets next and allows reflection, transmission, repeated encounters, and branching. It is widely used for illumination, light pipes, stray light, and ghost-image analysis. Statistical sampling can estimate how power spreads through a complicated enclosure, but an insufficient number of rays can miss weak yet important paths.
Optical design combines tracing with mathematical optimization. A merit function may penalize spot size, wavefront error, distortion, chromatic shift, length, mass, or sensitivity to fabrication errors. Variables include curvature, spacing, material, and surface coefficients. Tolerance analysis then evaluates departures such as element decenter, tilt, thickness error, and refractive-index variation.
A calculation is only as complete as its physical model. A geometrically valid path does not by itself predict rough-surface scattering, thin-film interference, or diffraction from a sharp edge. Engineering models therefore combine ray geometry with measured material data, coating models, scattering distributions, and wave propagation where needed.
Diffraction, caustics, and the limits of ray descriptions
Geometrical optics suppresses effects associated with finite wavelength. Diffraction spreads light near apertures and obstacles, while interference depends on relative phase between contributions that reach the same point. Independent power-carrying rays do not reproduce either phenomenon. A ray calculation can retain optical path for a later phase calculation, but this is an extension beyond a purely geometrical intensity model.
A circular aperture illustrates the difference. Geometrical optics can bring an on-axis bundle to a mathematical point, whereas a uniformly illuminated circular aperture produces an Airy diffraction pattern. Its first dark ring occurs at angular radius approximately
for small angles in air. In the focal plane of an imaging system, the corresponding radius is approximately . At wavelength and , this radius is about , even when geometrical aberrations vanish.
The geometrical approximation also fails at a caustic, an envelope of rays where neighboring trajectories concentrate. A familiar example is the bright curve formed when light reflects inside a cup onto its interior surface. Geometrical amplitude transport may predict divergent intensity because a ray-bundle cross-section tends to zero. Actual wave fields remain finite and typically show structured diffraction patterns. Away from the caustic, separate ray branches may again give useful asymptotic descriptions.
The transition between near-field and far-field diffraction can often be characterized using the Fresnel number,
where is a characteristic aperture radius and is propagation distance. Large can support a predominantly geometrical description in regions away from edges, but it does not make every interference feature disappear. The correct criterion depends on the observation region and the requested observable, not simply on whether an optical component is visibly larger than a wavelength.
Hybrid methods use ray tracing to find paths and wave propagation to determine the field around an image, aperture, or caustic. The point-spread function describes the intensity response to a point object under specified imaging conditions. It can incorporate both aberration and diffraction, providing information that a geometrical spot diagram alone cannot supply.
At a deeper level, the ray limit can be derived from Maxwell’s equations, while detection at low light levels may require the photon description of quantum mechanics. A ray is not a literal sequence of measured photon positions. It is an approximate propagation structure that remains useful across classical and quantum optical calculations when wavelength-scale and coherence effects are treated appropriately.
Historical development and theoretical unification
The mathematical study of rays began long before the wave nature of light was established. The ancient Greek optical tradition associated with Euclid treated visual directions geometrically and developed relations between apparent size, distance, and viewing angle. Such accounts often assumed rays extending from the eye, an interpretation distinct from the geometrical constructions themselves. In the second century CE, Ptolemy investigated reflection and refraction and compiled measurements of refracted angles, although his treatment did not yield the modern sine law.
In the early eleventh century, Ibn al-Haytham developed an account of vision based on light entering the eye from external objects. His Book of Optics combined geometrical analysis with experiments involving light, apertures, reflection, and refraction. The work’s later Latin transmission helped shape medieval European optics. Its treatment of image formation and visual perception nevertheless remained embedded in the anatomical and scientific knowledge of its time.
The development of spectacles, lenses, and astronomical instruments made practical image formation increasingly important. Johannes Kepler explained the formation of an inverted retinal image in his optical work of 1604 and analyzed lens combinations in Dioptrice in 1611. Willebrord Snell obtained the sine relation for refraction in 1621, and René Descartes published the relation in 1637. Pierre de Fermat subsequently showed how reflection and refraction could be understood through an extremal-time principle, providing a common variational foundation.
During the seventeenth century, ray reasoning coexisted with competing accounts of light’s physical nature. Christiaan Huygens developed a wavefront construction, while Isaac Newton used corpuscular ideas and extensive experiments on color. Newton’s demonstration that white light contains differently refracted colors was especially important for understanding chromatic aberration. Ray geometry alone did not settle the dispute between corpuscular and wave theories, because several basic geometrical laws could be reproduced within more than one physical interpretation.
Nineteenth-century developments placed the subject within a broader mathematical framework. William Rowan Hamilton developed characteristic-function methods that connected optical paths with ideas later central to Hamiltonian mechanics. Carl Friedrich Gauss’s 1841 work established a systematic first-order description of centered optical systems. Theories of aberration, improved optical glass, and precision fabrication then made increasingly sophisticated instruments possible.
The wave and electromagnetic theories of light clarified the domain in which rays are valid. James Clerk Maxwell unified optics with electromagnetism, and short-wavelength methods subsequently made the connection between field equations, eikonals, and ray trajectories explicit. Electronic computation in the twentieth century transformed ray tracing from laborious hand calculation into a central instrument-design technique, enabling optimization of systems containing many surfaces and detailed statistical analysis of manufacturing tolerances.