Reconceptualising Optical Resolution through Near-Field Interactions — Epoche C1
A grating ruled with a period of $200\,\mathrm{nm}$, illuminated by green light of wavelength $\lambda = 500\,\mathrm{nm}$ in air, sends no diffracted beam towards a distant lens: the information about its periodicity is present in the field immediately above its surface but has fallen to a thousandth of its amplitude within about fifty nanometres. This essay is about that fact and what follows from it. The diffraction limit is not a statement that fine detail is destroyed at the object; it is a statement about which components of the field can travel. The argument below derives the classical limits and shows exactly where their numerical factors come from, derives the decay of the components that carry sub-wavelength detail, describes what near-field microscopy measures, and then corrects a substantial error in the earlier version of this essay, which grouped the far-field super-resolution techniques STED and PALM/STORM with near-field methods. They do not capture evanescent waves at all, and understanding why sharpens the thesis rather than weakening it. Where the Classical Numbers Come From Two different criteria are in circulation and they give different numbers, so both derivations are worth having in front of us. Throughout, the numerical aperture $\mathrm{NA} = n\sin\alpha$ is the product of the refractive index $n$ of the medium between object and lens and the sine of the half-angle $\alpha$ of the cone of light the lens collects; it measures how wide a fan of directions the objective can gather. Abbe (1873) reasoned from a periodic object. A grating of period $d$ illuminated by light of wavelength $\lambda$ sends light into discrete diffracted orders at angles satisfying $n\sin\theta_m = m\lambda/d$. An image that shows the periodicity at all requires interference between at least two orders, since a single beam carries no spatial modulation. With the illumination along the axis, the objective must therefore collect both the zeroth and the first order, requiring $\mathrm{NA} \geq \lambda/d$, hence $d \geq \lambda/\mathrm{NA}$. If instead the object is illuminated at the steepest available oblique angle, the zeroth order enters at one edge of the collection cone and the first order at the opposite edge, so the angular budget is used twice over, and $$d_{\min} = \frac{\lambda}{2\,\mathrm{NA}}.$$ The factor of $2$ is therefore not a fudge: it is the gain from tilting the illumination, and it is lost if the condenser aperture is stopped down. Rayleigh (1879) reasoned instead from two self-luminous points. The image of a point source formed through a circular aperture is the Airy pattern, whose intensity is proportional to $[2J_1(v)/v]^2$, where $J_1$ is the Bessel function of the first kind of order one and $v = (2\pi/\lambda)\,\mathrm{NA}\,r$ is a dimensionless radial coordinate in the image plane. The first zero of $J_1$ beyond the origin lies at $v = 3.8317$. Rayleigh's convention declares two points just resolved when the centre of one Airy pattern falls on the first zero of the other, so setting $v = 3.8317$ gives $$R = \frac{3.8317}{2\pi}\cdot\frac{\lambda}{\mathrm{NA}} = \frac{0.61\lambda}{\mathrm{NA}}.$$ The $0.61$ is $3.8317/2\pi$ and nothing else. It differs from Abbe's $0.5$ because the two criteria answer different questions — one asks when a periodic structure is transferred, the other when two incoherent points are distinguishable — and because Rayleigh's is a convention about visual separability rather than a theorem. Neither number is a law of nature, which is the first hint that the barrier is softer than it is usually presented as being. Why Fine Detail Stops Propagating The deeper account replaces criteria with a decomposition, and it is here that the essay's central claim is established rather than asserted. Any field distribution across the plane $z = 0$ immediately above an object can be written as a superposition of plane waves — its angular spectrum — each labelled by its transverse wavevector components $k_x, k_y$. A plane wave of angular frequency $\omega$ in a medium of index $n$ must satisfy the dispersion relation $k_x^2 + k_y^2 + k_z^2 = k_0^2$, with $k_0 = 2\pi n/\lambda$. Writing $k_\parallel^2 = k_x^2 + k_y^2$ for the transverse part, each component acquires the factor $e^{ik_z z}$ on travelling a distance $z$ away from the object, where $$k_z = \sqrt{k_0^2 - k_\parallel^2}.$$ Two regimes follow directly, and the whole subject is contained in the difference between them. If $k_\parallel \leq k_0$, the square root is real, $e^{ik_zz}$ is a pure phase, and the component propagates without loss of amplitude. If $k_\parallel$ exceeds $k_0$, the square root is imaginary: writing $k_z = i\kappa$ with $\kappa = \sqrt{k_\parallel^2 - k_0^2}$ real and positive, the factor becomes $e^{-\kappa z}$. Such a component does not propagate away from the surface at all. It travels along the surface, and its amplitude falls exponentially with height. This is what is meant by an evanescent wave, and it corrects two loose statements in the earlier version: the decay constant is $\kappa$, not the wavevector $k$, and $k_z$ is not merely complex but purely imaginary, so there is no phase advance in $z$ whatever. Now connect transverse wavevector to feature size. A structure of period $\Lambda$ contributes a component with $k_\parallel = 2\pi/\Lambda$. That component is evanescent when $2\pi/\Lambda > 2\pi n/\lambda$, that is, when $$\Lambda < \frac{\lambda}{n}.$$ Every detail finer than roughly a wavelength in the medium is therefore encoded in a wave that cannot reach a distant lens. The classical resolution limits are not an additional physical postulate; they are this cutoff, expressed in the vocabulary of criteria. The decay is severe, and putting numbers to it explains why near-field probes must come so close. Take a feature of period $\Lambda = \lambda/10$ in air, so $n = 1$ and $k_\parallel = 20\pi/\lambda$. Then $$\kappa = \frac{2\pi}{\lambda}\sqrt{10^2 - 1} = 9.95\