The ray picture left a loose end that is about to become the whole story. Rays suggested that any zig-zag angle beyond the critical angle is guided — a smooth continuum of equally good paths. Measure a real waveguide and you find nothing of the sort: light propagates in a handful of discrete field patterns, sometimes exactly one, each moving at its own precise speed. Where did the continuum go? The answer is that a light ray is not a thin pencil; it is a wave, and a wave bouncing between two walls interferes with itself. Almost every zig-zag angle self-destructs. The few that survive are called modes, and the single number that prices each one — the effective index n_{\text{eff}} — will follow us through every device in this course. When a later lesson writes a phase as 2\pi n_{\text{eff}} L/\lambda without comment, it is leaning on today.

Why the angles are quantised

Follow one wavefront through a double zig-zag: down across the core, reflect, back up, reflect again. It has returned to its starting depth, travelling in its original direction — and it now overlaps the wavefronts of the undisturbed wave behind it. If the round trip has shifted its phase by anything other than a whole number of 2\pi, the overlap is destructive; a few more round trips and the field has cancelled itself to nothing. Exactly as in a standing wave on a string, persistence demands self-consistency. For a core of thickness d and a zig-zag angle \theta (from the normal), the transverse round trip covers 2d\cos\theta of path and picks up a phase shift 2\phi_r from the two total reflections, so the survivors obey

One equation, one integer — a discrete ladder of allowed angles \theta_m, each with its own transverse field pattern: m counts the nulls across the core. The continuum of rays has collapsed into a short list of modes, for precisely the reason electron orbitals or drum overtones form a short list: confinement plus wave interference equals quantisation.

Each allowed pattern travels down the guide as a wave E(x)\,e^{i(\beta z - \omega t)} with propagation constant \beta = (2\pi n_1/\lambda)\sin\theta_m — the axial component of the zig-zag. It behaves exactly like a plane wave in an imaginary uniform material of index

The brackets are worth internalising as a reflex. A well-confined mode lives mostly in the core and has n_{\text{eff}} near n_1; a barely-guided mode sprawls into the cladding and has n_{\text{eff}} sagging towards n_2. The effective index is a weighted average of the indices the mode's field actually touches — which is also why it responds to anything that changes those indices, a sensitivity that later lessons turn into switches, sensors and couplers.

The shape of a mode — and its evanescent tails

Solving the wave equation in matter for a symmetric slab gives the fundamental mode a shape you could have guessed: a cosine hump across the core. The surprise is at the walls. The field does not stop at the core boundary — it continues into the cladding as a decaying exponential, E \propto e^{-\gamma |x|}: the evanescent tail. These tails carry no power away (that is what makes the reflection total), but they are real, measurable field, typically reaching some 100–300 nm into the cladding of a silicon guide. The chart below is the exact solved mode of a symmetric slab; the slider sets the normalised guide size V = \tfrac{\pi d}{\lambda}\sqrt{n_1^2 - n_2^2}. Shrink V and watch the guide loosen its grip:

Two readings of the same picture. First, there is no single-mode "cliff" in confinement: even a comfortably single-mode guide keeps healthy tails, and a guide squeezed too small holds its mode so loosely that the merest bend sheds it. Second — and this is the seed of a lesson to come — the tail is a handle sticking out of the waveguide. Bring a second guide within tail's reach and light will climb across the gap, with no contact between the cores. The symmetric slab also hides a neat freebie: its fundamental mode has no cutoff at all. However tiny V gets, one mode always clings on; the second mode (m = 1) only appears once V > \pi/2 — and that inequality, reversed, is the single-mode condition.

Worked example: sizing a single-mode silicon guide

Put numbers in for the platform this course lives on: silicon (n_1 = 3.48) in silica (n_2 = 1.44) at the telecom wavelength \lambda = 1.55\ \mu\text{m}. The single-mode condition V < \pi/2 reads

A quarter of a micrometre — six times smaller than the wavelength in vacuum. High contrast is a double-edged sword: it permits micrometre bends, but it forces nanoscale cores if you want exactly one mode. (This is precisely why the standard silicon strip waveguide you will meet two lessons from now is about 0.5 µm × 0.22 µm — wide enough to hold the mode firmly, small enough to hold only one.) Compare telecom fibre, where \sqrt{n_1^2 - n_2^2} \approx 0.13: the same algebra allows a core about 6\ \mu\text{m} across — a size mismatch of more than 10× that will come back to haunt us when chip must meet fibre. The program below solves the actual mode equation — the transcendental self-consistency condition, not an approximation — and reports n_{\text{eff}} as the core shrinks:

Watch n_{\text{eff}} slide from near 3.5 down towards 1.44 as the core shrinks and the mode is squeezed out into the cladding — the bracketing inequality, live.

Where this goes next

You now hold the wave-level contract of a waveguide: a discrete set of modes, each a fixed transverse shape with a price tag n_{\text{eff}}, evanescent tails included. Photonic circuit design is largely the art of keeping every guide single-mode and then manipulating that one mode's phase. Before returning on-chip, though, the next lesson takes the waveguide idea to its greatest triumph — the optical fibre — and asks the question that decides whether guided light is useful at all: how much of it survives the trip?