The Abbe Diffraction Limit: Why a Light Microscope Can't See Atoms

Abbe's picture: an image is a diffraction reconstruction

Before Abbe, opticians thought a microscope simply projected a magnified shadow. Abbe realized that when light passes a fine object — say a grating of period d — it diffracts into discrete orders at angles set by the grating equation d·sin θ = m·λ. The zeroth order (m = 0) carries the average brightness; the ±1st and higher orders carry the fine structure. The objective lens gathers these beams and lets them re-interfere in the image plane to rebuild the object.

Here is the crux: a lens can only capture rays that fall within its acceptance cone. If the object is so fine that even the first diffracted order bends past the rim of the lens, the objective collects only the zeroth order — a featureless average — and the periodicity is invisible. To resolve period d you must capture at least the ±1st order, requiring

• the ±1st order angle θ to satisfy sin θ ≤ NA/n, i.e. the ray still enters the lens;
• which, via d·sin θ = λ, gives the smallest resolvable period d = λ / (n·sin θ) = λ / NA for one-sided (oblique) illumination;
• and d = λ / (2·NA) when the condenser floods the specimen from both sides (matched illumination), doubling the collectable angular range.

This is why Abbe treated resolution as an information-transfer problem, not a geometry problem: the lens is a low-pass filter on spatial frequency, and the cutoff is fixed by wavelength and aperture.

Numerical aperture: the only knob that matters

The numerical aperture bundles everything the lens can do: NA = n·sin θ, where n is the refractive index of the medium between specimen and lens and θ is the half-angle of the widest cone the objective accepts. It is dimensionless and it is the star of Abbe's formula.

• In air, n = 1.00 and sin θ < 1, so NA is capped below 1 — the best dry objectives reach NA ≈ 0.95.
• Immersing the front lens in oil with n ≈ 1.515 (matched to the coverslip) lets rays that would otherwise totally-internally-reflect at the glass–air interface reach the lens, pushing NA ≈ 1.4–1.5.
• Because sin θ ≤ 1 always, NA can never exceed n, and n for transparent immersion media tops out near 1.5. So NA is stuck around 1.5, and there is no more room.

Plug the best case into d = λ/(2·NA): with green light λ = 500 nm and NA = 1.4, d ≈ 500/(2·1.4) ≈ 179 nm. Halving that would require doubling NA to ~2.8, which is physically impossible in any real transparent medium. The famous ~200 nm textbook figure is simply this equation evaluated at visible wavelengths and the largest achievable NA.

The Fourier view: a lens is a low-pass filter

Modern optics restates Abbe in the language of spatial frequency. Any object can be decomposed into sinusoidal fringes of spatial frequency k = 2π/d (fine detail = high k). Propagating from object to lens, each spatial-frequency component leaves at a diffraction angle sin θ = k·λ/(2π). The finite aperture passes only components with

• k ≤ k_max = (4π·NA)/λ under two-sided illumination,
• which corresponds exactly to the smallest period d_min = 2π/k_max = λ/(2·NA).

Everything above k_max is lost forever — it never enters the imaging system, so no amount of computation on the recorded image can recover it. This is the honest reason a light microscope cannot see an atom: an atom's structure lives at spatial frequencies ~1000× above the optical cutoff, and those Fourier components carry no propagating wave the lens can catch. (They do exist as evanescent near-field waves that decay within ~λ of the surface — the loophole exploited by near-field microscopy, below.)

The transfer of the surviving frequencies is described by the optical transfer function (OTF), whose support is a disk of radius 2·NA/λ. Its magnitude, the MTF, falls to zero exactly at the Abbe cutoff — a curve every microscope and camera designer lives by.

Abbe, Rayleigh, and Sparrow: three limits, one physics

You will meet several 'resolution' numbers; they differ only in the prefactor and the imaging scenario, not in the underlying wave physics.

• Abbe (1873): d = λ/(2·NA). The finest resolvable grating period. The purest statement of the aperture cutoff.
• Rayleigh (1879): d = 0.61·λ/NA. Two self-luminous points are 'just resolved' when the peak of one Airy disk sits on the first dark ring of the other. The 0.61 comes from the first zero of the Bessel function J₁ (at 1.22π), giving sin θ = 1.22·λ/D for a circular aperture of diameter D.
• Sparrow: d ≈ 0.47·λ/NA. A slightly tighter criterion where the dip between two peaks just vanishes.

For NA = 1.4 and λ = 500 nm they give ≈ 179 nm (Abbe), ≈ 218 nm (Rayleigh), and ≈ 168 nm (Sparrow) — all within a factor of ~1.3 of each other. The lesson: the exact coefficient is a convention about 'how separated is separated,' while the λ/NA scaling is the law of nature. Any real microscope also demands Nyquist sampling — the detector pixel pitch (referred to the specimen) must be ≤ d/2 or the resolution is thrown away in aliasing, regardless of the optics.

Beating the wall: shorter waves and cleverer physics

Since d ∝ λ/NA, there are exactly two brute-force escapes, plus one elegant one.

• Shrink λ. Ultraviolet microscopes (λ ≈ 250 nm) roughly halve d to ~100 nm. Push further and you reach electron microscopes: a 200 keV electron has a de Broglie wavelength λ = h/p ≈ 2.5 pm — five orders of magnitude below visible light — letting a TEM resolve individual atomic columns at ≈ 0.05 nm. This is literally Abbe's formula with a matter wave (though lens aberrations, not diffraction, set the practical electron limit).
• Raise NA / use both sides. 4Pi and I⁵M microscopes use two opposed objectives to fill the aperture from more angles, sharpening the axial (depth) response several-fold.
• Cheat the far field entirely. STED (Stefan Hell, Nobel 2014) and PALM/STORM (Betzig, Moerner) do not raise NA at all — they use fluorescence switching so that only molecules farther apart than d ever emit at the same instant. STED narrows the effective spot as d/√(1 + I/I_sat), reaching 20–50 nm; single-molecule localization pinpoints one emitter to a few nm by fitting its Airy-disk centroid. These win by never asking the lens to transmit forbidden frequencies.

Near-field (NSOM) takes the last loophole: it probes the evanescent field within ~10 nm of the surface, before those high-k waves have decayed, reaching ~20 nm without any far-field diffraction limit at all.

Where the 200-nanometer wall shows up in the real world

The Abbe limit quietly governs an enormous swath of technology and biology.

• Cell biology: mitochondria (~500 nm) are easy; ribosomes (~25 nm), individual microtubules (~25 nm), and nuclear pores (~120 nm) sit at or below the limit, which is precisely why super-resolution earned a Nobel Prize.
• Photolithography: the same math sets the smallest transistor a chip can print. Node scaling drove the light from 365 nm (i-line) to 193 nm (ArF) and now 13.5 nm EUV; every wavelength cut is an Abbe-limit cut. Immersion lithography floods the gap with water (n ≈ 1.44) to raise NA past 1, and the industry's half-pitch = k₁·λ/NA is Abbe's formula with a process factor k₁.
• Astronomy: the circular-aperture cousin θ = 1.22·λ/D means the Hubble telescope (D = 2.4 m) resolves ≈ 0.058 arcsec — sharp only because D is 2.4 m, not because the optics are magic.
• Cameras and phones: stop a lens down to f/22 and the Airy disk swells past the pixel; images go soft. This diffraction-limited f-number is why more megapixels stop helping past a point.

Common misconceptions, sharpened

'Better glass would break the limit.' No. A perfect, aberration-free lens is exactly the diffraction-limited case — the limit is what remains after you fix every flaw. Abbe's number is the best case, not the worst.

'Magnification is resolution.' Magnify a 200 nm-limited image 10,000× and you get an empty magnification — a bigger blur, no new detail. Resolution is set by NA and λ; magnification just makes the resolved detail comfortable for the eye or sensor.

'You literally cannot locate anything smaller than 200 nm.' You cannot resolve two features closer than d, but you can localize a single isolated emitter to a precision ≈ d/√N by fitting its point-spread function with N collected photons — nanometers with enough light. Resolving many nearby objects and pinpointing one are different problems; super-resolution exploits exactly this gap.

'It's about the microscope, not the light.' The wall is intrinsic to propagating waves in a medium of index n. Any far-field imaging with waves of wavelength λ — sound, radar, light, electrons — obeys d ≈ λ/(2·NA). Only near-field tricks or matter waves with tiny λ get around it.

Resolution limits under different formulations and imaging modes (visible light, λ ≈ 500 nm unless noted)

Formulation / mode	Formula	NA or aperture	Resolvable detail d
Abbe (coherent, transmitted illumination)	λ / (2·NA)	NA = 1.4 (oil)	≈ 180 nm
Abbe (dry objective)	λ / (2·NA)	NA = 0.95 (air)	≈ 260 nm
Rayleigh (two points)	0.61·λ / NA	NA = 1.4 (oil)	≈ 218 nm
UV microscope	λ / (2·NA)	λ=250 nm, NA=1.25	≈ 100 nm
Electron microscope (TEM)	≈ 0.61·λ / NA	λ≈2.5 pm @ 200 kV	≈ 0.05–0.1 nm
STED super-resolution	d / √(1+I/I_sat)	NA=1.4, high I	≈ 20–50 nm