9.0 OPTICS, LENSES, AND ABERRATION CORRECTION⧉

The FUNDAMENTALS part told you a beautiful lie. The thin lens — a single ideal element of zero thickness for which $1/f = 1/u + 1/v$ — is what you draw on the whiteboard, and it is exactly right, as long as you keep only the first term of everything. It is the paraxial approximation: replace $\sin\theta$ by $\theta$ (drop the cubic term and all that follow), pretend there is only one wavelength, and pretend the glass has no thickness. Under those three pretences every ray from a point converges back to a point, and photography looks easy.

Reality keeps the terms the thin lens threw away, and the whole story of this part falls out of that one fact. Snell's law, $n_1\sin\theta_1 = n_2\sin\theta_2$, is non-linear: restore the next term, $\sin\theta \approx \theta - \theta^3/6$, and rays through the edge of the aperture no longer land where the paraxial rays do. The refractive index depends on wavelength ($n(\lambda)$, dispersion), so red and blue focus in different places. And real glass has thickness, real lenses have many surfaces, and real barrels have mounts, dust, and air gaps that scatter stray light. The consequence is stark: no single element can make all the rays, from all the field points, at all the wavelengths, through all of the aperture, converge to the right place. It is not a manufacturing failure; it is geometry.

So the lens you actually own is not one piece of glass — it is a system of six, ten, twenty elements, and it was found not by a formula but by optimization: vary the curvatures, glasses, thicknesses, and spacings to minimize the leftover blur, exactly the data-fit-versus-trade-off shape the rest of the book uses for inverse problems, here with a lens as the unknown and ray-tracing as the forward model. And whatever the glass cannot fix, computation calibrates and corrects — software undoes distortion, vignetting, color fringing, even some of the blur, from a measured profile of the lens. That partnership is the spine of the part. The optics only has to get close; the algorithm finishes the job. Optics and computation are two ends of one design space, and the most interesting modern cameras are designed across both at once.

The roadmap is eight chapters, and they share one throughline: where the thin lens breaks, and how we patch it — in glass, in mechanism, or in software.

The first three build the machine and explain why it has so many parts. Compound lenses teaches the lens as a black box with cardinal points — principal planes, entrance and exit pupils, the true f-number $N = f/D_{\text{EP}}$ measured at the pupil rather than the front element, the cinema T-stop that accounts for transmission loss — and tours the classic designs (the achromatic doublet, the Cooke triplet, the double-Gauss, telephoto versus retrofocus, the zoom) that each add elements to cancel a specific defect. Aberrations correction is the analytic heart: keep the cubic term and the five monochromatic Seidel aberrations appear (spherical, coma, astigmatism, field curvature, distortion); let $n$ depend on $\lambda$ and chromatic aberration appears. This is the catalogue of exactly how a real lens departs from the ideal point image, what each looks like in a photo, and the two complementary cures — in glass (aspheres, achromats, low-dispersion glass, stopping down) and in software (lens-profile correction, and deconvolution by the measured point-spread function (PSF)). Lens optimization then shows that lens design is optimization: you cannot solve "make all rays converge" in closed form past first order, so you minimize a merit function that sums residual blur over field × wavelength × aperture, evaluated by ray-tracing — the same forward-model-plus-loss-plus-iterative-solver machinery as everywhere else in the book, applied to glass, and the reason modern design runs on tools like Zemax OpticStudio and CODE V.

Special optics is a gallery of designs that deliberately break a thin-lens assumption to buy a capability: tilt the plane of focus (the Scheimpflug condition), abandon rectilinear projection for a 180°-plus fisheye, fold the path with mirrors (catadioptric, with its signature doughnut bokeh), squeeze the frame (anamorphic), image life-size (macro), and — the throughline to the Advanced part — trade the bulk of the glass for a thin structure, from the Fresnel lens through diffractive optics and gradient-index (GRIN) to flat metalenses.

The next two chapters are about the plane of sharpness and what surrounds it. Focus, autofocus treats focusing as choosing the conjugate that lands the subject on the sensor — mechanically trivial, but deciding which way to move is the real problem, solved by a ladder of methods: contrast-detection (climb the sharpness hill, but blind to direction), phase-detection autofocus (PDAF) (a split-pupil view that is literally stereo inside the lens, giving a signed defocus in one shot), its on-sensor and dual-pixel forms, and learned subject and eye detection on top. Depth of field then goes deeper on the blur itself — why an out-of-focus highlight takes the shape of the aperture (bokeh), why it clips to cat's-eye lemons toward the corners — and on extending it (focus stacking) or faking it (depth-aware portrait-mode blur), leaning on the FUNDAMENTALS geometry rather than re-deriving it.

The last two clean up what the rest leaves behind. Glare suppression is the mirror image of aberrations: aberrations are wanted rays in the wrong place, glare is unwanted light — inter-surface reflections and scatter — landing as ghosts, flare, and contrast-killing veiling glare. The cures split the familiar way: in hardware (anti-reflection coatings by thin-film interference, hoods, internal baffles) and in software (deflare and glare deconvolution). And optical stabilization fights the motion blur that hand-shake smears across an exposure: gyroscopes sense the shake, and a floating lens group (optical image stabilization, OIS) or the sensor itself (in-body image stabilization, IBIS) moves to keep the projected image still, buying several stops of shutter — with the honest caveat that it cancels camera motion only and cannot un-blur a moving subject, which is the deblurring chapter's job.

Eight chapters, one habit of mind. Watch the same lever recur: a defect that the thin lens hid, met either in glass — more elements, exotic surfaces, coatings — or in computation — calibrate, deconvolve, correct — and, increasingly, met by designing the two together. That last move, end-to-end or computational optics, is forward-referenced throughout to the Advanced part (coded apertures, wavefront coding, metalenses, lensless imaging), where the optics may look "wrong" by classical metrics yet give the better picture. Keep the partnership in view and the part stops looking like a catalogue of lens trivia and starts looking like one idea: in theory everything converges; in practice the glass gets close and the software does the rest.

Contents of this part

▸ full collapsible outline of this part

9.0 OPTICS, LENSES, AND ABERRATION CORRECTION🔗⧉

Contents of this part

9.0 OPTICS, LENSES, AND ABERRATION CORRECTION⧉