Cross two laser beams in mid-air and… nothing happens. Each sails through the other as if it were not there. This is not an accident of geometry — it is linearity, the same property this course has been quietly exploiting all along. Maxwell's equations in vacuum are linear, so light waves obey superposition: they add, they interfere, they redistribute — but one beam can never change another. Every device built so far — couplers, Mach–Zehnders, meshes, Fourier lenses — is a linear machine that shuffles amplitudes without ever letting one signal decide another's fate. Module 6 ran into this wall repeatedly: even the coherent Ising machine had to smuggle a nonlinear crystal into its loop to get any decision-making done. Digital logic is the wall at its highest. A logic gate is, by definition, a device in which one signal controls another — light must switch light. For that, superposition must break, and superposition only breaks where the medium responds nonlinearly. This lesson is about that response: where it comes from, what it can do, and — the number that will haunt the rest of this module — how absurdly hard you must drive matter to get it.

The anharmonic spring: P(E) beyond the linear term

Recall from electromagnetism in matter where a refractive index comes from: the optical field drives the electrons of the medium, the displaced electrons form a polarization P, and that polarization re-radiates, slowing and reshaping the wave. The textbook version keeps only the linear response, P = \varepsilon_0 \chi^{(1)} E — an electron on a perfect Hookean spring. But no real interatomic potential is a perfect parabola. Drive the electron hard enough and it explores the anharmonic part of its potential well, and its displacement stops being proportional to the push. Expand the response in powers of the field:

The scale of the series is set by the field that binds the electron in the first place: the atomic field E_{\text{at}} = e/(4\pi\varepsilon_0 a_0^2) \approx 5\times 10^{11}\ \mathrm{V/m}, the field a proton exerts one Bohr radius away. Each higher term in the expansion is suppressed by roughly a factor E/E_{\text{at}}. An intensity with an optical field equal to E_{\text{at}} would be about 3.5\times 10^{20}\ \mathrm{W/m^2} — at which point the material is not switching, it is exploding. Practical nonlinear optics lives many orders of magnitude below that, on the thin gruel of a slightly bent spring. That, in one sentence, is why every number in this lesson will be uncomfortably large.

χ⁽²⁾: the Pockels effect and frequency doubling

The second-order term mixes fields pairwise, and its two star turns are already familiar from earlier modules. Apply a DC field E_{\text{dc}} alongside the optical wave and the cross-term \chi^{(2)} E_{\text{dc}} E_{\text{opt}} acts like a change of linear susceptibility: the refractive index shifts linearly with the applied voltage. That is the Pockels effect, the physics inside every lithium niobate modulator — electronics steering light, fast and cheap. Feed in two optical fields instead (or one field with itself) and the E^2 term radiates at the sum frequency: second-harmonic generation, the trick that turns an infrared 1064\ \mathrm{nm} beam into the 532\ \mathrm{nm} green of a laser pointer.

Notice what χ⁽²⁾ offers a would-be logic designer: light controlling light, yes — but the product leaves at a new frequency. A gate whose output is a different colour from its input cannot feed an identical gate. Hold that thought; it becomes one of the sharpest knives in the criteria lesson.

χ⁽³⁾: the Kerr effect, by the numbers

For same-colour-in, same-colour-out control, the workhorse is third order. The \chi^{(3)} E^3 term includes a contribution oscillating at the original frequency whose strength rides on the local intensity — equivalent to a refractive index n = n_0 + n_2 I, with n_2 = 3\chi^{(3)} / (4 n_0^2 \varepsilon_0 c). A bright pulse raises the index under its own feet (self-phase modulation) and under any co-propagating signal (cross-phase modulation) — and a phase shift, as the Mach–Zehnder taught us, is one interferometer away from being an amplitude switch. So how big is n_2?

Ten to the minus twenty. To shift the index by a useful 10^{-4} in silica you need I = \Delta n/n_2 \approx 4\times 10^{15}\ \mathrm{W/m^2} — about ten billion times the intensity of full sunlight. The only reason Kerr switching is conceivable at all is that waveguides let you cheat on area: squeeze one watt into a silicon wire of cross-section 0.1\ \mathrm{\mu m^2} and the intensity is already 10^{13}\ \mathrm{W/m^2} — a gigawatt per square centimetre from a laser pointer's worth of power. The chart below prices the full switching operation: the phase shift \Delta\varphi = 2\pi n_2 I L/\lambda accumulated along a 1 cm waveguide of that cross-section, as a function of launched power. The horizontal line is \Delta\varphi = \pi — the full-switch threshold of an interferometer.

Read the silica curve carefully: it is not missing — it is flat. Even in the best nonlinear materials, a full switch costs of order a watt of optical power held inside a centimetre of waveguide, while the transistor it hopes to replace switches on femtojoules. This single chart is the seed of everything that goes wrong in the next three lessons.

Worked numbers: pricing a π

Let's make the chart's arithmetic explicit, because you will reuse it constantly in this module. For a waveguide of effective area A_{\text{eff}} and length L carrying power P:

Three sanity checks worth internalising. First, \Delta\varphi is linear in both intensity and length — doubling the waveguide is exactly as good as doubling the power, which is why fibre experiments happily use kilometres of silica instead of watts of power. Second, what matters is intensity, not power: shrinking A_{\text{eff}} is the cheapest nonlinearity-enhancer there is, and "make the mode smaller, or trap it longer" will be the entire design philosophy of the next lesson's devices. Third, silicon's 200× head start over silica comes with fine print: at 1.55 µm, two photons (0.8 eV each) jointly exceed silicon's 1.12 eV band gap, so two-photon absorption — itself a χ⁽³⁾ effect — burns off exactly the intense light you need, and dumps free carriers that linger for nanoseconds. The nonlinearity you want and the loss you don't are the same order of perturbation theory; you cannot order one without the other.

Where this goes next

We now have the raw ingredient: matter's grudging, intensity-bought willingness to let one light beam steer another — at gigawatts per square centimetre. The next lesson surveys forty years of ingenious machinery built to stretch that ingredient into working switches: semiconductor amplifiers that trade speed for sensitivity, fibre loops that trade footprint for it, and photonic-crystal cavities that trap light in a wavelength-sized box until even femtojoules feel intense. Keep today's two numbers in your pocket as you go: n_2 \sim 10^{-18}\ \mathrm{m^2/W}, and ten billion suns.