Optical parametric oscillator
A synchronously pumped singly resonant OPO in a bow-tie ring: green pump photons split into an 800 nm signal and a 1588 nm idler.
Open in the canvas →Background
An optical parametric oscillator makes new colours of coherent light without a laser transition. Inside a crystal with a χ⁽²⁾ nonlinearity, a pump photon splits into two lower-energy photons, the signal and the idler, whose frequencies add up to the pump's. Which pair appears is chosen by phase matching, so turning the crystal or changing its temperature tunes the output[1].
Energy conservation fixes the idler once the signal is chosen, and because the two are made in equal numbers of photons, the higher-energy signal carries the larger share of the generated power:
A single pass through the crystal gives little gain, so the crystal sits in a cavity that feeds one of the waves back. In a singly resonant OPO only that wave, here the signal, is reflected by the mirrors; the pump and idler pass straight through their coatings. Oscillation starts once the round-trip gain beats the round-trip loss, which is why an OPO has a pump threshold, and driven well above it the pump is depleted and the generated waves can start to convert back[2].
Here the cavity is a four-mirror bow-tie ring. In this singly pumped, co-propagating phase-matched configuration, parametric gain favours the signal travelling with the pump, selecting one circulation direction around the ring[1]. The mirror after the crystal is the output coupler: its coating reflects most of the signal and transmits the rest together with the pump and idler, so all three leave collinearly and are separated outside. A real ring resonator still requires a suitable mode, mirror curvatures, coatings and cavity length.
With a mode-locked pump the crystal only has gain while a pump pulse is inside it, so the resonator is synchronously pumped: its round trip takes exactly one pump period, and a signal pulse that leaves the crystal comes round to meet the next pump pulse and is amplified again[1]. For a ring that fixes the perimeter, c / frep — strictly the group optical path, so a dispersive crystal makes the real ring slightly shorter.
What this setup demonstrates
A 532 nm, 2 ps, 80 MHz pump drives a χ⁽²⁾ crystal between M1 and M2, the lower pair of a bow-tie ring completed by M3 and M4. All four are band reflectors for the 650–950 nm signal band and transmit the pump and idler, and every one is met at a 12° angle of incidence. The signal runs M2 → M3 → M4 → M1 → crystal, crossing itself between the upper and lower pairs. The perimeter is exactly c / 80 MHz = 3747.4 mm, a 12.5 ns round trip equal to the pump period; the crystal is a thin surface, so the whole path is air. Because of that, the animated packets of successive round trips overlap instead of forming offset clusters. The schematic view still draws several illustrative packets around the ring at a fixed fraction of the pulse spacing; the physical view shows the true pump-period spacing. The overlap follows from the length ratio, not from a simulated synchronisation or gain process. M2 is the output coupler: it reflects 80 % of the signal band and transmits the rest, so signal, idler and residual pump leave together; a longpass and a shortpass dichroic then separate them. Beam probes read the wavelength of the pump, the signal circulating on the upper arm, and each separated output: idler, signal and residual pump.
The crystal is set to a 532 nm pump and an 800 nm signal, so the inspector shows the idler at 1588 nm. It converts a fixed 30 % of the pump: 19.95 % becomes signal and 10.05 % idler. The idler detector reads that 10.05 % in one pass. The signal leaks through M2 on every round trip, and the tracer sums twelve of those leaks, reading 18.6 % of the pump. Those are successive round trips of one launched signal, not amplified passes: the tracer does not add gain from later pump pulses. It also launches the signal only in the pump's direction and does not calculate the gain competition that selects that direction. The signal is authored with the same frequency (wavenumber) FWHM as the pump, a heuristic, which is about 0.47 nm at 800 nm; the idler width is derived from pump and signal, and the output pulses keep the pump's duration; all are illustrative, not one instrument's.
Things to try: move the signal wavelength and watch the idler follow; lower M2's in-band reflectivity and more of the signal leaves on each round trip; or tune the pump more than 1 nm away from 532 nm and the oscillator goes dark.
This is a phenomenological OPO. The crystal converts a fixed fraction of the pump on its first pass: there is no threshold, gain, build-up or back-conversion, and phase matching is not calculated, so turning or heating the crystal does nothing. The signal output is a finite sum of traced leaks, not a steady state, and changing the ring's length does not detune anything — in a real synchronously pumped OPO it shifts the signal and can stop oscillation. Ring cavities usually use curved mirrors to focus into the crystal; these four are drawn flat. The pump is a single axial ray, so only the chief-ray routing is shown, with no mode, focus or beam overlap. Mirror coatings switch perfectly at their band edges and do not depend on angle.