Harnessing Evanescent Waves in Near-Field Optics — Epoche C1
A conventional microscope cannot show two objects 100 nanometres apart as two objects, and the reason is not that light is too coarse a probe: it is that the part of the scattered field which carries that information never travels far enough to reach the lens. Within a few tens of nanometres of the specimen the information is present and strong. It simply falls off exponentially with distance instead of propagating, and a lens placed at any ordinary working distance receives none of it. Near-field optics is the practice of collecting that field before it dies, and this essay sets out where the cut-off comes from, how fast the discarded part decays, and what it costs to catch it. One claim in the earlier version of this essay should be withdrawn at the outset. It described the diffraction limit as commonly and mistakenly attributed to "light's wave-particle duality". Duality has nothing to do with the matter, and the limit is not a quantum statement at all: it follows from classical Maxwell theory alone, by a two-line argument given below, and it would hold in exactly the same form for sound or for water waves. What the earlier text was reaching for is correct and more interesting — the limit is a constraint on propagation, not on light-matter interaction, and it therefore applies only to information that has been allowed to propagate. Where the diffraction limit comes from The derivation rests on one idea: any field distribution in a plane can be written as a superposition of plane waves, and each of those plane waves has its fate decided independently. Take the field in the plane $z=0$ just above a specimen and decompose it into components varying as $\exp\!\left[i\left(k_x x + k_y y\right)\right]$. The pair $(k_x, k_y)$ is the transverse spatial frequency : large values correspond to fine detail, since a component with transverse wavenumber $k_\parallel = \sqrt{k_x^2+k_y^2}$ describes a pattern of period $\Lambda = 2\pi/k_\parallel$. Each such component propagates into $z > 0$ as a plane wave, and Maxwell's equations in a medium of refractive index $n$ require every plane wave to satisfy $$k_x^2 + k_y^2 + k_z^2 = n^2 k_0^2, \qquad k_0 = \frac{\omega}{c} = \frac{2\pi}{\lambda_0},$$ where $\lambda_0$ is the vacuum wavelength. Solving for the component along the optical axis, $$k_z = \sqrt{n^2 k_0^2 - k_\parallel^2}.$$ Here the argument turns, and everything else follows from the turn. If $k_\parallel \le n k_0$, then $k_z$ is real and the component travels: it carries energy to a distant lens with its amplitude undiminished. If $k_\parallel$ exceeds $n k_0$, the quantity under the square root is negative, $k_z$ becomes purely imaginary, and writing $k_z = i\beta$ turns the propagation factor $e^{ik_z z}$ into $e^{-\beta z}$ — a field that does not travel but decays, with $$\beta = \sqrt{k_\parallel^2 - n^2 k_0^2}.$$ These are the evanescent components. They are not a correction or an approximation; they are exactly half of the solution set, and they are precisely the half that encodes detail finer than the wavelength. A lens intercepts only rays arriving within its acceptance angle $\theta$, so it collects components up to $k_\parallel = n k_0 \sin\theta = \mathrm{NA}\cdot k_0$, where $\mathrm{NA} = n\sin\theta$ is the numerical aperture. The finest grating period a microscope can transmit under illumination from a single direction is therefore $\Lambda_{\min} = 2\pi/(\mathrm{NA}\,k_0) = \lambda_0/\mathrm{NA}$. Abbe's criterion, quoted in the earlier version as $\lambda/(2\,\mathrm{NA})$, carries an extra factor of two whose origin is worth stating because it is routinely omitted. Illuminating the specimen obliquely rather than along the axis shifts every object frequency by the transverse wavenumber of the illuminating beam. If the condenser has the same numerical aperture as the objective, an object frequency as high as $2\,\mathrm{NA}\,k_0$ can be shifted down into the collection cone and detected. The factor of two is the sum of an illumination aperture and a collection aperture, not a property of light. Hence Abbe's 1873 result for the smallest resolvable separation, $$d = \frac{\lambda_0}{2\,\mathrm{NA}},$$ which for an oil-immersion objective of $\mathrm{NA} = 1.4$ and green light at $\lambda_0 = 500\ \mathrm{nm}$ gives $d \approx 179\ \mathrm{nm}$. How fast the discarded information decays The evanescent components exist; the question is how long they last, and this is the calculation that fixes every engineering requirement in near-field microscopy. For detail much finer than the wavelength, $k_\parallel \gg n k_0$, so the $n^2k_0^2$ term under the square root is negligible and $$\beta \approx k_\parallel = \frac{2\pi}{\Lambda}, \qquad\text{so that}\qquad \frac{1}{\beta} \approx \frac{\Lambda}{2\pi}.$$ This is the central result of the subject in one line: the distance over which the information about a feature of size $\Lambda$ survives is about $\Lambda/6$, and it depends on the feature size, not on the wavelength. The finer the detail one wants, the closer one must get, in strict proportion. The table below evaluates the exact expression for $\lambda_0 = 500\ \mathrm{nm}$ in air, listing the amplitude decay length $1/\beta$ and the fraction of amplitude surviving at a probe height of $10\ \mathrm{nm}$. Feature period $\Lambda$ $1/\beta$ (amplitude decay length) Amplitude at $z = 10\,\mathrm{nm}$ $200\ \mathrm{nm}$ $34.7\ \mathrm{nm}$ 0.75 $100\ \mathrm{nm}$ $16.2\ \mathrm{nm}$ 0.54 $50\ \mathrm{nm}$ $8.0\ \mathrm{nm}$ 0.29 $20\ \mathrm{nm}$ $3.2\ \mathrm{nm}$ 0.043 $10\ \mathrm{nm}$ $1.6\ \mathrm{nm}$ 0.0019 Two consequences follow directly. First, the earlier statement that the probe must sit "within a few nanometers of the sample surface" is now derived rather than asserted, and it is derived per target resolution: 20-nanometre detail requires the probe within roughly 3 nanometres if the signal is not to fall below a twentieth of its surface value. Second, since the detecte