A single quantum dot in a waveguide emits its photon in both directions at once — left and right with equal probability, because the waveguide is symmetric and the dot has no reason to prefer one side. That symmetry is the problem this paper solves. A team at the University of Copenhagen reports, in PRX Quantum on 16 November 2026, that two coupled quantum dots in a photonic-crystal waveguide emit photons in a direction that can be chosen: set the relative phase between two excitation lasers one way and the photons go left; set it another way and they go right. The control is the first demonstration of directional single-photon emission at optical frequencies. The microwave version already exists — superconducting qubits have been made to emit directional microwave photons — but microwaves are overwhelmed by thermal photons at room temperature, which limits their use in a network. Optical photons are the carriers a quantum network would actually use. The work comes from Clara Henke, Thomas Wilkens Sandø and colleagues at the Center for Hybrid Quantum Networks (Hy-Q) at the Niels Bohr Institute, led by Cornelis van Diepen.
The molecule and the waveguide
The waveguide is the first thing to get right, because everything in the experiment follows from its geometry. The team etched a photonic-crystal pattern — an array of holes with one row missing — into a suspended gallium-arsenide membrane that acts as a p-i-n diode. The missing row is the waveguide, with grating couplers at each end carrying photons out of the chip. The dots sit inside the waveguide, and the two that form the pair are 13 micrometres apart — a distance the paper notes corresponds to 26 effective wavelengths, with the wavelength at the photonic-crystal band edge being 500 nanometres, twice the 250-nanometre lattice constant. Thirteen micrometres is far for interacting dots; the coupling is mediated by the spatially extended guided mode, so the dots talk through the light they share.
That only works if the two dots emit at the same frequency, and dots grown on a wafer do not. They vary in size, and size sets frequency. The team’s solution is electrical: each dot’s emission frequency is tuned independently through the dc Stark effect, using local electric fields applied through separate electrodes. A shallow trench etched through the p-doped layer isolates the left and right halves of the waveguide electrically, so the two dots can be tuned without crosstalk. A transmission measurement confirms the independence: each dot’s resonance can be moved without moving the other’s.
The dots are strongly coupled to the guided mode. A transmission experiment makes this concrete: a weak continuous-wave laser at 320.861 terahertz is sent into the left port and collected at the right, while each dot is tuned in and out of resonance. On resonance, a single dot acts as a mirror; the transmission dips to 0.492(4) of the input for the first dot and 0.426(4) for the second, which the authors read as evidence that both dots’ β-factors — the fraction of their emission going into the guided mode — are high, close to unity. With both dots on resonance, the dip deepens to 0.239(3) — two dots reflect more efficiently than one. It is also shallower than the product of the two individual dips, 0.210(3), and the combined dip sits at a small detuning and is asymmetric in the individual detunings — signatures, the paper argues, of a dispersive component in the coupling, which shifts the energies of the collective states. The measured total decay rates are Γ₁/2π = 0.388(1) gigahertz and Γ₂/2π = 0.349(1) gigahertz.
The dispersive component is what distinguishes this experiment from earlier coupled-dot work. The coupling between two waveguide-coupled emitters has two parts: a dissipative part, which changes the decay rates and produces super- and subradiance, and a dispersive part, which shifts the energies of the collective states. Both depend on the coupling phase φ = kx, the phase lag accumulated between the emitters, where x is their separation and k the effective wave number of the guided mode. In earlier work on radiatively coupled dots in a photonic-crystal waveguide, the coupling was predominantly dissipative — φ was a multiple of π — and the emission was symmetric. Here the dots sit about 3.5 terahertz from the band edge, at 317.4 terahertz, and are separated by roughly 50 unit cells; the combination gives a coupling phase of φ = (0.8 + N)π, a regime with a substantial dispersive component, confirmed by simulations that reproduce the data at φ = 0.8π. Because the energy levels of the collective states are shifted, the pair effectively forms an artificial molecule — two dots and a waveguide behaving as one object with its own level structure.
Steering by phase
With the molecule established, the first control experiment is the headline one. Both dots are driven from free space with pulsed resonant excitation, using a spatial light modulator to split a single laser beam into two beams with controlled relative positions, driving strengths and phase. The drive is kept weak, so the emission comes predominantly from the single-excitation subspace.
The direction then depends on the relative phase θd between the driving fields: for the collective state (|eg⟩ + e^{iθd}|ge⟩)/√2, the left- and right-going fields are proportional to 1 + e^{i(θd + φ)} and 1 + e^{i(θd − φ)}. Set θd = π − φ and the left-going field cancels; set θd = π + φ and the right-going field vanishes. The measurement does exactly this: as the phase is swept, emission switches from bidirectional to predominantly rightward, then predominantly leftward. The paper’s wording is “predominantly” — the measured directionality is strong but not the idealised complete cancellation of the theory. The control is real and reversible; the perfection is asymptotic.
The second experiment steers without touching the driving phase: only the left dot is excited, with a calibrated π-pulse, while the undriven dot’s frequency is tuned. As it comes into resonance, the time-resolved emission at both ports is modulated — direct evidence of radiative coupling — with an asymmetry between ports and detuning signs that dissipative-only work did not show. In the time-integrated emission over 2 nanoseconds, the dominant direction switches from left to right as the detuning changes sign. Detuning, like driving phase, is a knob on the emission direction.
Single photons one way, pairs the other
The third experiment is the one that makes the result more than a routing demonstration. Under continuous, weak resonant driving of the left dot only, the team measures the second-order correlation g² at each port. With the undriven dot far detuned, both ports show ordinary single-emitter antibunching: g²_LL(0) = 0.13 and g²_RR(0) = 0.2, both well below the 0.5 single-photon threshold. Bring the undriven dot into resonance and the two ports diverge strikingly. On the left, the dip becomes shallower, g²_LL(0) = 0.31 — still antibunched, still predominantly single photons. On the right, an antidip appears: g²_RR(0) = 0.98, a bunching peak where a dip used to be. Single photons leave to the left; photon pairs leave to the right.
The mechanism explains why the asymmetry is not an artefact. In the near-unity-β limit, the single-excitation component of the steady state becomes the state |π + φ⟩, from which emission goes only to the left, while the double-excited state can decay in either direction. So the left port receives single photons plus a share of pairs, while the right port receives pairs only — antibunching on one side, bunching on the other. The undriven dot acts as a saturable mirror: it reflects single photons back toward the left, but saturates when two photons arrive, letting the pair through to the right. Driving both dots collectively gives the same control: depending on the relative driving phase, the team measures either a bunching antidip (g²_LL(0) = 1.05, g²_RR(0) = 1.71) or antibunching at both ports (0.40 and 0.39).
The final experiment turns the pair into an on-demand source of correlated photon pairs. Both dots are simultaneously driven with pulsed resonant excitation at π-flip power for each, fully inverting the pair into the doubly excited state |ee⟩. Emission from |ee⟩ is a two-step cascade: the first photon takes the system to one of the single-excitation states, and the second photon’s direction depends on which, so the two photons’ directions are temporally correlated. Across all four direction combinations, the normalised centre-peak heights with both dots resonant are g²_LL(0) = 0.70, g²_LR(0) = 0.42, g²_RL(0) = 0.41 and g²_RR(0) = 0.76 — a clear same-port versus different-port contrast. For comparison, exciting only the second dot gives 0.01, 0.05, 0.02 and 0.02: ordinary single-photon antibunching. The paper calls this emergent chirality — directionality arising from the spontaneously broken mirror symmetry of the coupled pair, rather than from any chirality built into the waveguide. A fully time-resolved correlation map confirms the picture: two photons from the same pulse arrive together, in a feature absent for photons from different pulses — a cascaded decay, or equivalently one emitter stimulating the other.
The honest boundary
Two qualifications are the standard ones for this kind of physics. The device operates in a closed-cycle 4 kelvin cryostat, and the dots emit at about 934 nanometres — near-infrared, but not a telecom wavelength, and not matched to the fibre infrastructure a deployed quantum network would use. Neither is a flaw, but both constrain any deployment story.
The third qualification is about what “directional” means here. The measured emission is predominantly to one side, not exclusively, and the phase control that produces it comes from a spatial light modulator shaping free-space beams above the chip — an external, laboratory-scale control, not an on-chip one. Increasing the directionality would require on-chip phase shifters and a secondary coupling mechanism that cancels the dispersive energy-level shifts while preserving the interference. Until those exist, the device demonstrates the control principle; it is not a routable component.
The fourth is scale. This is two emitters, selected from several in the waveguide because they happened to be nearly the same size and therefore tunable into resonance. Scaling to more emitters faces the two familiar obstacles: the random spatial locations of self-assembled dots and the inhomogeneous broadening of the ensemble. Simulations, not a demonstration, address this: sets of three and four resonant dots are already feasible with the present tuning range, the paper argues, and sets of ten could follow from a moderate few-fold increase in relative tunability and emitter density. Those are credible routes; none is demonstrated here.
None of these qualifications diminishes the result; they locate it. What Henke, Sandø and colleagues have done is take the collective-emission physics the microwave community demonstrated with superconducting qubits and realise it at optical frequencies, with local electrical control over each emitter and photon statistics showing the directionality extends to the quantum character of the light itself. The waveguide is bidirectional and non-chiral; the chirality is emergent, engineered into the state of the emitters rather than into the structure. That is the conceptual point: a symmetric, scalable waveguide platform can host directional quantum optics without needing a chiral structure at all.
What to watch next
The follow-ups that would matter most are easy to name, because the paper names them. The first is the on-chip phase shifter: moving phase control from a free-space spatial light modulator onto the chip, turning a laboratory demonstration into a circuit component. The second is the gate: conditional phase gates and deterministic photon sorting — the applications that would justify calling the device an optical switch rather than a demonstration of one. The third is spin: operating multiple dots with a spin degree of freedom, toward deterministic spin–spin entanglement and two-dimensional cluster states for photonic quantum computing. The fourth is the scaling test: three and four resonant dots in one waveguide, which the simulations say is already feasible.
A single quantum dot emits in both directions because it has no reason to choose. Two dots, coupled through a waveguide and driven with the right relative phase, do have a reason — and for the first time at optical frequencies, the reason can be dialled in from the outside. The photon still goes where the interference tells it to. Now the interference can be told where to send it.
Sources and further reading
- Clara Henke, Thomas Wilkens Sandø, Vasiliki Angelopoulou, Lena Maria Hansen, Alexey Tiranov, Oliver August Dall’Alba Sandberg, Zhe Liu, Leonardo Midolo, Nikolai Bart, Arne Ludwig, Anders Søndberg Sørensen, Peter Lodahl & Cornelis Jacobus van Diepen, “Directional and Correlated Optical Emission from a Waveguide-Engineered Molecule with Local Control,” PRX Quantum 7, 043032 (2026). DOI: 10.1103/v1r7-lr89. Published 16 November 2026. Primary source for all figures in this article. Preprint: arXiv:2604.06410 (submitted 7 April 2026).
- Michael Schirber, “Quantum Dots Emit Photons in Desired Directions,” Physics 19, 155 (16 November 2026). https://physics.aps.org/articles/v19/155 — the APS Focus summary; used for the framing of the microwave precedent and the optical-switch application.
- Prior microwave directional-emission work cited in the paper: Aamir et al., Physical Review Letters 129, 123604 (2022), and related superconducting-qubit results — the precedent this work extends into the optical domain. Prior optical work cited: directional scattering from a pair of silicon vacancies in a microdisk resonator, which lacked selective driving to control the direction.
- Prior waveguide-coupled-dot work from the same group: super- and subradiance in radiatively coupled dots in a photonic-crystal waveguide, with predominantly dissipative coupling (φ = Nπ) — the symmetric regime this paper moves beyond.
- Note on verification: all quantitative values here are taken from the published PRX Quantum paper and its arXiv v1 text, which match the published abstract; the full text was read in the arXiv HTML rendering. The emission wavelength of about 934 nanometres is computed from the paper’s stated laser frequency of 320.861 terahertz. No market-size figure, commercial claim, or investment judgement appears in this article; the site’s Terms carry the not-investment-advice disclaimer.