Abbe Theory of Image Formation

Abbe's back focal plane is the workbench of modern optical processing. A mask there filters spatial frequencies directly: a central stop gives dark-field imaging, a quarter-wave phase dot on the zeroth order gives Zernike phase contrast that renders transparent cells visible, matched filters enable optical pattern recognition — all pure Fourier-plane operations that rays cannot explain. The same decompose-filter-recombine logic underlies the $4f$ correlator and the Optical Transfer Function. Photolithography lenses are engineered to the Abbe limit, which is why the industry pushed to immersion (raising $n$, hence NA) and to shorter wavelengths. The limit Abbe set is also the bar super-resolution had to clear. Because $d_{\min}=\lambda/2\,\mathrm{NA}$ follows from which orders the pupil admits, far-field fluorescence methods that exploit photophysics rather than pupil geometry — STED, PALM and STORM, recognised by the 2014 Nobel Prize — reconstruct detail an order of magnitude finer, while Fourier ptychography synthesises a larger effective NA from many illumination angles. Each is best understood as an extension of Abbe's spectrum, not a repeal of it.

Put a fine grating on the microscope stage. Coherent light strikes it and splits into an undiffracted central beam plus $\pm1,\pm2,\dots$ orders fanning out at $\sin\theta_m=m\lambda/d$ — the finer the grating, the wider the fan. The objective funnels only the beams inside its acceptance cone; once the grating is fine enough that even the first order misses the lens, no interference occurs and the image is a uniform blur. That is resolution death made visible, and it is the whole of Abbe's insight in one observation.

The back focal plane is where the optics performs a Fourier transform for free: each object spatial frequency maps to a distinct diffracted spot, so blocking a spot deletes that frequency from the image — the basis of dark-field and phase-contrast microscopy. The single deepest fact is that two beams are the minimum for a fringe: one plane wave has uniform intensity and encodes nothing, so the lens must collect at least two orders and resolution has a hard cutoff, not a gentle fade. How that cutoff moves with numerical aperture is the punchline of the next figure.

Wider aperture, shorter wavelength, higher-index immersion all do the same thing: they let the lens catch steeper diffracted beams — higher spatial frequencies — so the low-pass filter passes more of the object's Fourier content and the image sharpens. That is why the numerical aperture, not the magnification, sets what a microscope can actually see.

Write the object as a superposition of sinusoidal gratings — its spatial-frequency spectrum $T(\nu)$. Each component is one grating that will diffract the illumination in a definite direction, and fine detail lives at high spatial frequency.

where: $t$ — complex transmittance of the object; $T$ — its Fourier spectrum; $\nu_x,\nu_y$ — spatial frequencies (cycles per unit length). The resolution question is how much of $T$ the lens can transmit. Each component grating of period $d$ diffracts the illumination into orders whose angles obey the grating equation of Grating Equation — the finer the grating, the larger the angle.

where: $\theta_m$ — angle of the $m$-th order; $d$ — grating period; $\lambda$ — wavelength in the medium; $m$ — integer order. The zeroth order carries no detail; the information about $d$ lives in the direction of the $\pm1$ orders. A lens gathers these plane waves and, in its back focal plane, sorts them by direction into position: the field there is the Fourier transform of the object, so each order appears as a spot — the transforming property developed in Introduction to Fourier Optics.

where: $U_f$ — field in the back focal plane; $x_f,y_f$ — coordinates there; $f$ — focal length; spectrum sampled at $\nu=x_f/\lambda f$. Insert a stop or phase plate there and you edit the image frequency by frequency. The pupil has a finite radius, so it passes only frequencies below a cutoff — a low-pass filter whose cutoff is fixed by the numerical aperture.

where: $P$ — coherent pupil function; $\nu_c$ — cutoff frequency; $\mathrm{NA}=n\sin\theta$; $n$ — object-space index; $\theta$ — acceptance half-angle. Every frequency above $\nu_c$ is discarded outright, so fine detail is lost abruptly, not gradually. Transforming the truncated spectrum back to real space, the image is the object convolved with the inverse transform of the pupil — the amplitude point-spread function, which for a circular pupil is the Airy amplitude of Circular Aperture Diffraction and the Airy Pattern.

where: $U_i$ — image amplitude (unit magnification); $h$ — amplitude point-spread function; $*$ — 2-D convolution; $r$ — image-plane radius. Truncating the spectrum is the same as smearing the object with $h$ — the origin of all diffraction blur. Resolution follows from demanding that at least the zeroth and one first order survive the pupil; with the extra factor of two from oblique illumination, this is Abbe's limit.

where: $d_{\min}$ — smallest resolvable period; $\mathrm{NA}=n\sin\theta$. The two-beam requirement makes the cutoff hard; oblique illumination doubles the reach by letting the zeroth and first orders enter at opposite edges of the cone.

How the Abbe limit responds to illumination and medium (NA = numerical aperture, $n$ = object-space index).

1 · The oil-immersion diffraction limit (confidence builder)

An oil-immersion objective has $\mathrm{NA}=1.4$ and works at $\lambda=550$ nm. Find the resolvable period (oblique and on-axis), the pupil cutoff, and the point-spread-function radius.

Step 1. Oblique (full condenser): $d_{\min}=\lambda/2\,\mathrm{NA}=550/(2\times1.4)=$ 196 nm — Abbe's own limit, the figure on his Jena memorial and the target that shaped Zeiss objective design.

Step 2. On-axis coherent: only the zeroth order is on the axis, so the limit doubles to $d=\lambda/\mathrm{NA}=550/1.4=$ 393 nm.

Step 3. Pupil cutoff: $\nu_c=\mathrm{NA}/\lambda=1.4/(550\times10^{-9})=2.545\times10^{6}\ \text{m}^{-1}=$ 2545 cyc/mm; every object frequency above this is discarded.

Step 4. Point-spread-function radius: the Airy first zero sits at $0.61\,\lambda/\mathrm{NA}=0.61\times550/1.4=$ 240 nm, the two-point (Rayleigh) radius. So one objective yields 196 nm (grating, oblique), 240 nm (two points), and 393 nm (grating, on-axis) — all from the single number NA = 1.4.

2 · The Abbe–Porter experiment: deleting orders deletes the image (exam level)

A grating of period $d=1\ \mu\mathrm{m}$ is imaged at $\lambda=550$ nm with the $\mathrm{NA}=1.4$ objective. Which orders form the image, and what happens if the $\pm1$ orders are blocked in the back focal plane?

Step 1. Diffraction angles: $\sin\theta_1=\lambda/d=0.550$, so $\theta_1=\arcsin(0.550)=$ 33.4°; the second order needs $\sin\theta_2=1.10>1$ and does not exist, so the grating gives only the 0 and $\pm1$ orders.

Step 2. Capture: the acceptance is $\sin\theta\le\mathrm{NA}/n=1.4/1.515=0.924$, and $0.550<0.924$, so the objective easily collects the $\pm1$ orders — the grating is well resolved (its period far exceeds the on-axis limit of 393 nm).

Step 3. Block the $\pm1$ spots in the back focal plane, leaving only the zeroth order. A single plane wave reaches the image, its intensity is uniform, and the grating disappears into a flat grey field — Porter's 1906 demonstration that the image needs at least two orders.

Step 4. Block instead the zeroth order (a central stop): only the $\pm1$ orders survive and beat at twice the object frequency, giving a bright, edge-enhanced dark-field image. Editing the back focal plane edits the image. The image is only ever as complete as the set of orders the pupil, and any masks in it, allow through.

The whole theory turns on one quantity — the numerical aperture. It is not a property of the object or the wavelength but of the lens: the sine of its acceptance half-angle, scaled by the index of the medium in front of it. It alone decides which diffracted orders enter, hence which spatial frequencies survive, hence how fine the detail the microscope can render.

Texts. Hecht, Optics, 5th ed., §13.2 (Abbe theory, spatial filtering) · Goodman, Introduction to Fourier Optics, 3rd ed., Ch. 5–6 · Saleh & Teich, Fundamentals of Photonics, 3rd ed., Ch. 4 and Ch. 24 · Born & Wolf, Principles of Optics, 7th ed., §8.6 · Loudon, The Quantum Theory of Light, 3rd ed. Key relations used throughout: the Fourier-transforming property of a lens follows from cascaded Fresnel integrals with quadratic-phase cancellation for a front-focal-plane object; the grating equation $\sin\theta_m=m\lambda/d$ and the two-beam conditions yield $d=\lambda/\mathrm{NA}$ (on-axis) and $\lambda/(2\,\mathrm{NA})$ (oblique); the convolution theorem gives image = object * PSF with the circular-aperture PSF $\propto 2J_1/(k\,\mathrm{NA}\,r)$; and the numerical resolution is $\approx0.18\,\mu\text{m}$ for $\lambda=0.5\,\mu\text{m}$, $\mathrm{NA}=1.4$. The anchor set throughout is 196 nm, 393 nm, 2545 cyc/mm, and 240 nm (NA = 1.4, λ = 550 nm).

Prerequisites. Fraunhofer Diffraction · Huygens–Fresnel Principle · Fresnel–Kirchhoff Diffraction Integral · Fourier transform. Leads to. Rayleigh Resolution Criterion · Circular Aperture Diffraction and the Airy Pattern · Fourier Optics and Spatial Filtering · Optical Transfer Function.