Quantum Optics Frontiers

Status: antibunched emission, heralded single photons, two-photon interference, optical squeezing, Gaussian continuous-variable processing, non-Gaussian state preparation, integrated quantum-photonic components, and remote matter-memory entanglement are established. High-efficiency indistinguishable sources, loss-tolerant integrated circuits, optical Gottesman–Kitaev–Preskill resources, practical quantum repeaters, and fault-tolerant photonic architectures are active. A general-purpose quantum internet, a fault-tolerant optical quantum computer, and task-independent quantum advantage from present photonic processors are not established.

Last reviewed: 26 July 2026. Source efficiencies, squeezing records, network rates, loss budgets, classical simulation methods, and fault-tolerance thresholds are date-sensitive. A component benchmark conditional on detecting a photon is not an end-to-end system benchmark.

Core Question

How can nonclassical optical states be generated, processed, transmitted, stored, and measured at a scale where the claimed quantum resource survives the complete experiment?

Quantum optics has exceptionally mature primitives. A beam splitter can implement a precise mode transformation. A local oscillator can select a quadrature. Optical fibre can carry a telecom photon over kilometres. Parametric interactions can create squeezing and photon pairs. Individual emitters can produce antibunched light. None of those facts alone solves a system problem.

A useful optical experiment is a chain:

$$\begin{aligned} \text{source} &\longrightarrow \text{mode preparation} \longrightarrow \text{circuit or channel} \\ &\longrightarrow \text{memory or feed-forward} \longrightarrow \text{detector} \longrightarrow \text{inference}. \end{aligned}$$

Loss, distinguishability, phase drift, multiphoton contamination, detector noise, memory decoherence, and postselection enter at different arrows. Their effects are not interchangeable. A source may have excellent $g^{(2)}(0)$ yet poor extraction efficiency. A circuit may have high conditional fidelity while accepting very few trials. A network may distribute secret keys without distributing stored entanglement between quantum processors.

The frontier is therefore resource-preserving systems engineering with claim-level validation. The central questions are:

1. Which optical state and mode carry the resource?
2. Which operation, channel, or measurement consumes it?
3. Which loss and noise parameters determine accepted-event probability?
4. Which witness or task verifies that the resource survived?
5. Which scale variable actually improved: rate, mode count, fidelity, distance, integration density, stability, or logical performance?

Why It Matters

Photons are natural carriers

Optical photons propagate quickly, interact weakly with ordinary thermal environments, and can occupy polarization, path, time-bin, frequency, and spatial modes. Those properties make them natural flying systems for communication and natural links among otherwise incompatible matter platforms.

Weak environmental coupling is also a limitation. A photon that is easy to transmit is difficult to store, route conditionally, or make interact deterministically with another photon. Most optical architectures replace direct photon–photon interactions with measurement, ancillary states, matter interfaces, or nonlinear processes.

Quantum optical measurements are unusually expressive

Photon counting accesses discrete number statistics. Homodyne detection accesses field quadratures with a mode-selective local oscillator. Heterodyne detection samples both quadratures with added vacuum noise. Interferometers convert phase and mode coherence into count or photocurrent statistics.

The same field can therefore be interrogated in complementary ways. This supports precise tests of state models, but it also means that phrases such as “single photon,” “squeezed mode,” and “entangled light” are incomplete until the temporal, spectral, spatial, and polarization modes are declared.

Photonics spans discrete and continuous encodings

A dual-rail photonic qubit uses one excitation across two modes. A continuous-variable system uses quadratures of one or many bosonic modes. Time and frequency multiplexing can supply many addressable modes without a separate macroscopic path for every mode.

These encodings have different resource bottlenecks. Dual-rail schemes pay heavily for loss and probabilistic entangling operations. Gaussian continuous-variable schemes create large entangled states deterministically, but universal fault-tolerant computation requires a non-Gaussian ingredient and finite-squeezing error correction.

Squeezing is both foundational and operational

Squeezed light demonstrates that vacuum fluctuations can be redistributed between conjugate quadratures. It is also deployed in precision interferometry. Frequency-dependent squeezed-vacuum injection now improves full-scale gravitational-wave detectors over a broad frequency band.

This is a useful standard for the rest of the field: the resource is not merely generated and reconstructed in isolation. It changes the performance of a scientific instrument under an explicit loss and noise budget.

Integrated photonics can turn alignment into fabrication

Free-space experiments offer access and flexibility, but every additional optic creates alignment, phase, vibration, and packaging burdens. Integrated waveguides, resonators, phase shifters, sources, and detectors can replace some of those burdens with lithographic reproducibility and electronic control.

Integration does not make loss disappear. Coupling interfaces, crossings, filters, switches, thermal tuning, fabrication variation, detector temperature, and control wiring become system constraints. Wafer-scale fabrication is a capability; high-yield operation of a complete quantum system is a stronger capability.

Optical networks connect otherwise local quantum systems

Matter qubits can store and process information while photons carry it. Frequency conversion can bridge an emitter wavelength to a low-loss telecom band. Heralding can turn uncertain transmission into a known successful event.

A practical network must combine entanglement generation, memory lifetime, local gates, conversion, transmission, detection, synchronization, and classical control. Demonstrating one link is an enabling result, not yet a multi-hop repeater network.

Canonical Boundary

This page evaluates current capabilities, system metrics, and open research questions. It does not duplicate stable derivations.

• Quantum Optics owns the operational field language and the map of states, transformations, and measurements.
• Quantized Electromagnetic Modes owns mode quantization, normalization, and creation and annihilation operators.
• Photon-Number States owns Fock states and number statistics.
• Correlation Functions and Antibunching own coherence functions and nonclassical count statistics.
• Beam Splitters owns linear mode transformations and Hong–Ou–Mandel interference.
• Parametric Down-Conversion and Nonlinear Quantum Optics own pair generation, phase matching, and nonlinear interactions.
• Squeezed Light owns squeeze operators, quadrature spectra, generation, calibration, and loss.
• Homodyne and Heterodyne Detection owns optical receiver models and mode selection.
• Phase-Space Distributions owns optical quasiprobabilities and tomography.
• Photonic Qubits owns discrete photonic encodings, source-to-detector acceptance, linear-optical processing, cluster and fusion architectures, feed-forward, and fault-tolerance resource ledgers.
• Continuous-Variable Systems and Gaussian States: Preview own covariance matrices, symplectic structure, and Gaussian-state entanglement.
• Gaussian Channels and Erasure and Loss Channels own channel maps, capacities, and open-system interpretation.
• Cavity QED and Input–Output Theory Overview own controlled light–matter interfaces and propagating-field boundary conditions.

The present page owns dated comparisons such as whether a reported “on-demand” source is actually deterministic at the system output, whether a network stores entanglement or only distributes keys, and whether a non-Gaussian state has reached a stated fault-tolerance target.

Minimal Background

A photon belongs to a mode

For a normalized spectral-temporal mode $f(\omega)$,

$$\hat a_f^\dagger = \int d\omega\, f(\omega) \hat a^\dagger(\omega), \qquad \int d\omega\,|f(\omega)|^2=1.$$

The full mode also includes spatial profile and polarization. A state that contains one excitation in mode $f$ is

$$|1_f\rangle = \hat a_f^\dagger|0\rangle.$$

Two sources can emit exactly one photon per pulse and still fail to interfere if their modes differ. Conversely, filtering can improve mode overlap while reducing the accepted rate. Source quality is therefore multidimensional.

A source is a metric vector

A useful source report contains at least

$$\mathcal M_{\mathrm{src}} = \left( P_1,\, g^{(2)}(0),\, \eta_{\mathrm{ext}},\, \mathcal I,\, B,\, f_{\mathrm{rep}},\, S \right).$$

Here:

• $P_1$ is the probability of occupying the desired one-photon sector in the declared source mode;
• $g^{(2)}(0)$ diagnoses equal-time number correlations and, with a source model, constrains multiphoton contamination;
• $\eta_{\mathrm{ext}}$ is extraction or delivery efficiency to a declared reference plane;
• $\mathcal I$ is indistinguishability in the relevant interference test;
• $B$ is useful brightness or accepted count rate;
• $f_{\mathrm{rep}}$ is repetition rate; and
• $S$ summarizes stability, tuning range, and duty cycle.

No one coordinate substitutes for the others. A low $g^{(2)}(0)$ at a tiny collection efficiency is not a high-rate source. A bright source with spectral wandering is not automatically useful for interference among independent emitters.

For a stationary field,

$$g^{(2)}(0) = \frac{ \langle \hat a^\dagger \hat a^\dagger \hat a \hat a \rangle }{ \langle \hat a^\dagger\hat a \rangle^2 }.$$

An ideal one-photon Fock state has $g^{(2)}(0)=0$. Detector dead time, background subtraction, temporal binning, and blinking can bias an estimate, so the reported value needs a measurement model.

Pair sources trade probability against multipair noise

An ideal two-mode squeezed-vacuum pair source can be written

$$|\psi\rangle = \sqrt{1-q} \sum_{n=0}^{\infty} q^{n/2} |n\rangle_s|n\rangle_i, \qquad 0\le q<1.$$

The pair-number distribution and mean are

$$P_n = (1-q)q^n, \qquad \mu = \frac{q}{1-q}.$$

Conditioned only on the event $n\ge1$, the ideal multipair fraction is

$$P(n\ge2\mid n\ge1) = q = \frac{\mu}{1+\mu}.$$

Increasing pump strength raises the heralding probability but also raises the multipair fraction. Finite herald efficiency, dark counts, spectral multimode structure, and number resolution modify the conditional state. This is why “turn up the pump” is not a free source-rate improvement.

Two-photon interference measures mode overlap

For distinguishable and nominally indistinguishable settings with coincidence counts $C_{\mathrm{dist}}$ and $C_{\mathrm{ind}}$, a common raw Hong–Ou–Mandel visibility is

$$V_{\mathrm{HOM}} = \frac{ C_{\mathrm{dist}}-C_{\mathrm{ind}} }{ C_{\mathrm{dist}} }.$$

For ideal single photons with density operators $\rho_1$ and $\rho_2$ and an otherwise ideal receiver,

$$V_{\mathrm{HOM}} = \operatorname{Tr}(\rho_1\rho_2).$$

Real visibilities also depend on multiphoton events, beam-splitter imbalance, detector timing, background, polarization, and data corrections. Raw and background-corrected values answer different questions and should both be identified.

Squeezing is a calibrated variance ratio

For

$$\hat X_\theta = \frac{ \hat a e^{-i\theta} + \hat a^\dagger e^{i\theta} }{ \sqrt2 },$$

the vacuum variance is $V_{\mathrm{vac}}=1/2$ in this convention. A quadrature is squeezed when $V_\theta<V_{\mathrm{vac}}$. The reported noise reduction is often

$$S_{\mathrm{dB}} = -10\log_{10} \left( \frac{V_\theta}{V_{\mathrm{vac}}} \right).$$

A pure-loss channel with transmissivity $\eta$ gives

$$\frac{V_{\mathrm{out}}}{V_{\mathrm{vac}}} = \eta \frac{V_{\mathrm{in}}}{V_{\mathrm{vac}}} + (1-\eta).$$

Even perfect vacuum loss injects fluctuations. Correcting a measured variance back to a source plane can be informative, but an application sees the uncorrected variance at its operating reference plane.

Gaussian processing is powerful but closed

For $N$ modes, collect quadratures into

$$\hat{\mathbf R} = \left( \hat q_1,\hat p_1,\ldots,\hat q_N,\hat p_N \right)^{\mathsf T}.$$

A Gaussian state is characterized by a mean vector $\mathbf d$ and covariance matrix

$$V_{jk} = \frac12 \left\langle \left\{ \hat R_j-d_j,\, \hat R_k-d_k \right\} \right\rangle.$$

Gaussian unitaries act as

$$\mathbf d \longmapsto S\mathbf d+\mathbf s, \qquad V \longmapsto SVS^{\mathsf T},$$

where $S$ is symplectic. Beam splitters, phase shifts, displacements, and squeezers belong to this family. Gaussian channels add loss and noise. Homodyne and heterodyne receivers are Gaussian measurements.

These operations support teleportation, sensing, communication, and large cluster states. Gaussian states processed only by Gaussian operations and measurements remain within a class that is efficiently tractable under standard descriptions. Universal quantum computation therefore needs a non-Gaussian state, operation, or measurement.

Non-Gaussianity must be defined operationally

A state can be called non-Gaussian because its Wigner function is not Gaussian. Wigner negativity is a stronger witness than mere non-Gaussianity, but it is not the only possible resource criterion. Photon subtraction, photon addition, number-resolving conditioning, strong light–matter interactions, and encoded grid states can supply non-Gaussian resources.

For quadratures satisfying $[\hat q,\hat p]=i$, ideal square-lattice Gottesman–Kitaev–Preskill code states are stabilized by

$$\hat S_q = e^{i2\sqrt{\pi}\hat q}, \qquad \hat S_p = e^{-i2\sqrt{\pi}\hat p}.$$

They are non-normalizable combs in the ideal limit. Physical states have finite peak width and a finite-energy envelope. Their usefulness depends on the effective shift-error distribution after preparation, loss, refinement, gates, and measurement, not merely on whether several peaks are visible.

Fibre loss is exponential in distance

For attenuation $\alpha$ in $\mathrm{dB/km}$ and length $L$ in kilometres,

$$\eta_{\mathrm{ch}} = 10^{-\alpha L/10}.$$

At $\alpha=0.2\ \mathrm{dB/km}$, $100\ \mathrm{km}$ gives $\eta_{\mathrm{ch}}=10^{-2}$. A direct protocol requiring two independently transmitted photons can inherit a factor $\eta_{\mathrm{ch}}^2$ if each photon traverses the full distance.

A heralded elementary-link probability is more honestly written as

$$\begin{aligned} p_{\mathrm{link}} \approx{}& p_{\mathrm{emit}}\, \eta_{\mathrm{couple}}\, \eta_{\mathrm{convert}}\, \eta_{\mathrm{ch}}\, \eta_{\mathrm{det}} \\ &\times p_{\mathrm{protocol}}\, p_{\mathrm{accept}}. \end{aligned}$$

The mean waiting time is set by this probability and the attempt cycle. Useful memories must preserve the relevant state while neighbouring links are established, classical heralds return, and swapping or purification is performed.

Multiplexing moves rather than removes cost

If $M$ independent source attempts each succeed with probability $p$, ideal multiplexing gives

$$P_{\ge1} = 1-(1-p)^M.$$

A binary selection network has at least $d=\lceil\log_2 M\rceil$ switching stages. If each stage transmits a photon with probability $\tau$, a selected photon carries a factor

$$\eta_{\mathrm{switch}} = \tau^d.$$

Multiplexing also requires delay, feed-forward, synchronization, and collision handling. A useful comparison includes all of them.

Integrated performance needs a resource ledger

For a circuit with components $j$, a first loss budget is

$$\eta_{\mathrm{path}} = \prod_j \eta_j.$$

That product does not contain phase errors, thermal crosstalk, fabrication yield, detector saturation, reset time, control latency, cryogenic power, or calibration overhead. A system ledger should therefore report:

• loss from source plane to detector;
• conditional operation fidelity;
• accepted-event probability and rate;
• mode count and connectivity;
• detector efficiency, dark counts, jitter, and recovery time;
• phase stability and recalibration interval;
• component and circuit yield across dies or wafers;
• feed-forward latency and memory or delay depth; and
• resources used to discard or correct unsuccessful events.

The best value measured on each component need not have been achieved simultaneously in one operating system.

Evidence levels answer different questions

This page uses five labels.

• Established: repeatedly demonstrated with an operational definition and a measurement model.
• Active: supported by current experiments or theory, with important scaling or integration questions unresolved.
• Conjectural: a quantitatively motivated extrapolation not yet demonstrated at the claimed scale.
• Speculative: a possible direction without a validated end-to-end path.
• Controversial: experts disagree about an inference, benchmark, or comparison, not merely about terminology.

An optical state can be established while its use in a scalable architecture remains conjectural.

What Is Known

Antibunched emitters can produce highly indistinguishable photons

Established. Resonantly driven semiconductor quantum dots in engineered photonic environments have produced bright, low-multiphoton, highly indistinguishable single-photon streams. The 2013 demonstration by He et al. and the 2016 near-optimal source by Somaschi et al. were decisive milestones.

The phrase “near-unity indistinguishability” refers to a declared Hong–Ou–Mandel experiment under specific excitation, filtering, time separation, and correction conventions. It does not imply unit extraction, perfect long-term mutual indistinguishability among many remote devices, or lossless delivery to a processor.

Pair sources remain workhorses

Established. Spontaneous parametric down-conversion and spontaneous four-wave mixing provide photon pairs, heralded single photons, entangled pairs, and squeezed vacuum across bulk, fibre, waveguide, and resonator platforms. They are compatible with telecom wavelengths and integrated fabrication.

Their probabilistic pair statistics are not a defect in characterization; they are the physical source model. Weak pumping suppresses multipair events while reducing herald rate. Number-resolving detection, spectral engineering, and multiplexing improve the trade, but do not erase its resource cost.

Independent solid-state emitters can interfere on chip

Established at small scale. In 2025, an integrated molecular platform placed 100 waveguide-coupled molecules within Stark-tuning reach of a common frequency. Two independently coupled molecules showed on-chip Hong–Ou–Mandel visibility above $0.97$, and controlled detuning produced quantum beating lasting beyond $100\ \mu\mathrm{s}$.

This establishes independent-emitter mode matching and long optical coherence for the measured pair. It does not establish simultaneous hundred-photon interference or a processor made from 100 operating sources.

Squeezing can survive a complete scientific instrument

Established. Direct detection of $15\ \mathrm{dB}$ squeezed vacuum at $1064\ \mathrm{nm}$ was reported in 2016 under a carefully calibrated laboratory loss budget. In full-scale LIGO detectors, frequency-dependent squeezing using $300\ \mathrm{m}$ filter cavities reduced both shot noise and radiation-pressure noise over a broad band.

The 2023 LIGO report found noise-amplitude reductions of $4.0\ \mathrm{dB}$ near $1\ \mathrm{kHz}$ at Hanford and $5.8\ \mathrm{dB}$ at Livingston. The low-frequency improvement increased detector range by roughly $15\%$–$18\%$ relative to no squeezing, corresponding to as much as a $65\%$ increase in expected detection rate under the stated scaling.

The source record and instrument improvement are different metrics. The latter includes injection loss, mode mismatch, phase noise, filter-cavity loss, and the detector response.

Continuous-variable teleportation and cluster generation are real

Established. Unconditional optical continuous-variable teleportation was demonstrated in 1998. Deterministic temporal multiplexing later generated a two-dimensional cylindrical cluster state containing more than 30,000 entangled modes, with 24 modes around the circumference and 1,250 temporal steps.

Those results establish large Gaussian entangled resources and measurement-based processing primitives. Finite squeezing introduces noise at each teleportation step. A Gaussian cluster state without a suitable non-Gaussian resource and fault-tolerant encoding is not by itself a universal fault-tolerant computer.

Non-Gaussian optical states are established

Established. Conditional photon subtraction has generated Schrödinger-cat-like optical states with Wigner negativity. Photon counting and nonlinear light–matter interactions have prepared many other non-Gaussian states. Optical non-Gaussianity is therefore not merely a proposal.

What remains active is the combination of high state quality, high probability, multiplexability, low-loss processing, and error correction at architectural scale.

Integrated optical GKP structure has been observed

Established as state generation; active for fault tolerance. A 2025 silicon-nitride experiment generated propagating optical GKP qubit states using an integrated Gaussian-boson-sampling circuit and high-efficiency photon-number-resolving detection. The reconstructed states showed at least four resolvable peaks in each quadrature and a $3\times3$ lattice of negative Wigner-function regions.

The authors’ fault-tolerance analysis concluded that further optical-loss reduction is required for devices of this type to produce states in the target fault-tolerant regime. Visible grid structure is a major source milestone, not evidence that a complete optical logical processor already suppresses errors.

Matter memories have been entangled through telecom fibre

Established for elementary links. In 2024, two nanophotonic silicon-vacancy nodes used spin–photon gates, telecom frequency conversion, and heralding to entangle remote spin memories. The experiment used up to $40\ \mathrm{km}$ of fibre spool and a $35\ \mathrm{km}$ deployed urban loop with $17\ \mathrm{dB}$ loss. Nuclear-spin storage reached the one-second scale with error-detection tools; remote electron-spin entanglement reached rates up to the hertz scale under the relevant settings.

This is an elementary two-node memory network and an enabling repeater component. It is not an end-to-end demonstration of a multi-link repeater beating direct transmission.

Multi-emitter nodes can supply multiplexing

Established at proof-of-principle scale. A 2025 rare-earth-ion experiment used multiple distinguishable ytterbium ions in nanophotonic cavities. Frequency-erasing detection and feed-forward generated entanglement between remote ion pairs, demonstrated multiplexed rate enhancement with two pairs, and prepared a three-ion $W$ state.

This directly addresses the one-emitter-per-node bottleneck. The next questions are yield, crosstalk, optical stability, memory coherence, and scaling the number of simultaneously useful emitters.

A memory can outlive mean link establishment

Established for a 2026 trapped-ion link. Remote ion memories connected by $10\ \mathrm{km}$ of spooled fibre retained entanglement longer than the average time required to establish it. The same platform supported a proof-of-principle device-independent key-distribution analysis at $10\ \mathrm{km}$ and an asymptotic positive-key projection over $101\ \mathrm{km}$.

Crossing the lifetime-to-waiting-time boundary is a critical repeater milestone. A useful repeater still needs multiple elementary links, entanglement swapping, a full finite-rate analysis, and comparison with a direct-transmission benchmark.

Integrated photonics can co-fabricate core discrete-variable functions

Established as a component stack. A 2025 commercial $300\ \mathrm{mm}$ silicon-photonics process integrated pair sources, filters, interferometers, and superconducting detectors. Reported benchmarks included:

Function	Reported benchmark	Essential scope
dual-rail state preparation and measurement	$99.98\%\pm0.01\%$ fidelity	conditional on photon detection
independent-source interference	$99.50\%\pm0.25\%$ visibility	selected operating conditions
two-qubit fusion	$99.22\%\pm0.12\%$ fidelity	conditional; loss excluded
chip-to-chip interconnect	$99.72\%\pm0.04\%$ fidelity	conditional; loss excluded

The paper explicitly states that these values do not account for loss and that the heralded sources are nondeterministic. It separately reports next-generation low-loss components, switches, and detectors. Combining the best numbers into a hypothetical machine is an architecture projection, not a measured end-to-end benchmark.

Gaussian photonic processors have reached large sampling regimes

Established for specific sampling tasks. Borealis used 216 squeezed modes, dynamic time-bin interferometers, and photon-number-resolving detection for Gaussian boson sampling. Events contained up to 219 detected photons, and the experiment argued a large runtime separation from the best classical methods then considered.

In 2026, Jiuzhang 4.0 reported 1,024 squeezed inputs, an 8,176-mode spatial–temporal circuit, $92\%$ source efficiency, $51\%$ overall system efficiency, and detected events up to 3,050 photons. These are remarkable large-scale optical sampling experiments.

They establish neither universal computation nor a useful speedup for an arbitrary application. Classical simulation algorithms continue to improve, especially by exploiting loss, graph structure, limited correlations, or approximate objectives. Every advantage statement belongs to a specified distribution, error model, validator, and classical baseline.

What Is Actively Developing

Deterministic and multiplexed single-photon sources

The target is not simply one photon after a trigger. It is a photon in a declared mode, delivered at a useful reference plane, with low multiphoton probability, high mutual indistinguishability, high repetition rate, and long-term stability.

Active approaches include:

• resonantly driven quantum dots in microcavities and waveguides;
• colour centres and rare-earth ions coupled to nanophotonic resonators;
• organic molecules with Stark tuning;
• heralded pair sources with active temporal, spectral, or spatial multiplexing;
• atom–cavity emission and collective atomic interfaces; and
• frequency conversion into a common telecom mode.

The meaningful comparison is an accepted-photon rate at fixed purity and indistinguishability, including switching and coupling loss.

Source stabilization and autonomous calibration

Integrated resonators drift with temperature, pump power, and fabrication variation. A 2025 electronic–photonic system-on-chip in a commercial $45\ \mathrm{nm}$ CMOS process stabilized several photon-pair microrings with local feedback under thermal disturbances.

That is an important control result. Stabilizing a probabilistic source does not make it deterministic. The frontier is to scale calibration and feedback without consuming excessive optical power, detector bandwidth, chip area, or cryogenic cooling.

Low-loss switching and routing

Multiplexed sources, fusion architectures, and feed-forward measurements need fast switches. A switch must be evaluated through insertion loss, extinction, phase stability, voltage or energy per operation, bandwidth, footprint, crosstalk, and compatibility with the source and detector temperature.

A one-percent loss looks small for one switch. Repeated across many stages and photons, it becomes a dominant erasure probability.

Scalable non-Gaussian resource factories

Photon subtraction and number-resolved conditioning are effective but probabilistic. Optical GKP preparation converts a difficult non-Gaussian state-synthesis problem into a resource-factory problem: many candidate outputs may be generated, characterized, refined, and routed into a cluster.

The active questions are:

• raw state probability;
• finite-energy envelope and effective shift noise;
• loss before and after heralding;
• detector efficiency and number resolution;
• correlations among accepted states;
• refinery overhead;
• magic-state quality; and
• logical error after a complete gate or teleportation cycle.

Continuous-variable fault tolerance

Large Gaussian cluster states can be generated deterministically. The remaining challenge is to combine finite squeezing, loss, non-Gaussian resources, feed-forward, and decoders so that increasing code size lowers a logical error rate.

GKP encodings are prominent because Gaussian Clifford operations can act deterministically on encoded modes and small displacement errors can be digitized. The threshold depends on the complete noise model. Quoting an equivalent squeezing number without preparation loss, correlated errors, decoder assumptions, and circuit depth is incomplete.

Monolithic continuous-variable circuits

In 2026, a monolithic circuit integrated squeezed-light generation, single-mode and two-mode Gaussian gates, local oscillators, interferometers, and balanced homodyne detection. It prepared and verified four-mode cluster states on chip.

This closes an important integration loop at small mode count. Scaling must retain squeezing, phase stability, local-oscillator quality, detector linearity, and controllability while adding modes and non-Gaussian capabilities.

Hybrid discrete–continuous architectures

Photon counting can condition non-Gaussian states from Gaussian light. Homodyne measurements can process GKP-encoded information. Matter emitters can supply nonlinearity or storage while optical modes supply transport.

Hybrid designs are attractive because each subsystem addresses another’s weakness. They are difficult because mode conversion and interface loss can erase the nominal benefit. The correct comparison is architectural overhead at fixed logical target, not the elegance of one primitive.

Practical quantum repeaters

Elementary-link demonstrations now combine memory, telecom conversion, and heralding. Active repeater work focuses on:

• memory coherence exceeding stochastic link waiting times;
• temporal, spectral, and spatial multiplexing;
• high-fidelity local gates and Bell measurements;
• entanglement swapping and purification;
• asynchronous scheduling;
• finite-key and finite-rate analyses;
• deployed-fibre stability; and
• comparisons with repeaterless bounds.

An experiment becomes a repeater demonstration when it uses intermediate quantum resources to improve a declared end-to-end task, not merely when its apparatus contains a memory.

Integrated quantum communication

A 2026 laboratory network combined one silicon-nitride microcomb server with 20 indium-phosphide QKD transmitter chips. It sequentially performed pairwise twin-field QKD through ten wavelength channels, each corresponding to $370\ \mathrm{km}$ of spooled fibre, and reported key rates above the repeaterless bound under the protocol’s network geometry.

This establishes impressive photonic integration, wavelength multiplexing, and multi-client QKD control. It is not an entanglement-swapping network and does not provide general quantum-state transport among processors.

Matter–photon interfaces and transduction

Network nodes must connect stationary memories to low-loss optical modes. Active interfaces include cavity-enhanced atoms, ions, quantum dots, colour centres, rare-earth ions, atomic ensembles, and microwave-to-optical transducers.

For transduction, conversion efficiency and added noise must be reported together. A 2026 experiment coherently transferred signals between superconducting circuits in separate dilution refrigerators over $1\ \mathrm{km}$ of telecom fibre. The authors described it as a path toward a quantum-enabled link; preservation of arbitrary quantum states and end-to-end quantum-channel performance remain stronger tests.

Application-level continuous-variable processing

Continuous-variable photonics is active in sensing, communication, sampling, and machine learning. A 2026 reservoir-computing experiment used deterministically generated multimode squeezed states, spectral and temporal multiplexing, mode-selective homodyne detection, and electro-optic feedback to control fading memory and perform temporal tasks.

This establishes a reconfigurable quantum optical reservoir. It does not by itself establish a computational advantage over optimized classical reservoirs. A fair advantage test must match data access, trainable parameters, energy, sampling cost, and error bars.

Reproducible system benchmarks

The field increasingly needs benchmark suites that travel across laboratories and platforms. Useful suites should include raw data, calibration records, accepted-event definitions, uncertainty propagation, source and detector reference planes, adversarial classical models, and machine-readable circuit descriptions.

The hardest benchmark is often not a larger state. It is a result whose claim survives a change of detector, calibration team, or classical analysis.

What Is Debated or Uncertain

What counts as on demand

An emitter excited by a trigger may emit into the desired mode only with probability $\eta_{\mathrm{src}}<1$. A heralded source announces successful generation after a detector click. A multiplexed source may deliver a photon with high probability after active routing.

All three are useful, but they are different. A defensible “on-demand” claim states the trigger-to-delivery probability, reference plane, multiphoton error, timing window, and whether heralding or postselection is used.

Whether conditional fidelity is the right headline

Conditional fidelity asks whether an accepted photon or event is correct. Loss asks how often an event is accepted. In an erasure-tolerant architecture these can be separated to some degree, but fault tolerance still imposes an erasure threshold and overhead.

A $99.9\%$ conditional operation with a $1\%$ acceptance probability and a $99\%$ acceptance operation with lower fidelity may serve different architectures. Neither number ranks them alone.

Whether a source record transfers to a system

Source brightness is often measured near the emitter. Detector efficiency may be measured on a separate die. Switch loss may be measured with classical light. Circuit fidelity may be conditioned on detection.

Combining independently optimized values is useful for a design model, but it is not a demonstrated system. Uncertainty and correlation among component parameters also matter.

Whether a QKD network is a quantum internet

QKD distributes correlated secret classical keys. An entanglement network distributes nonseparable quantum states. A repeater stores and swaps entanglement across elementary links. A distributed quantum computer also requires remote logical operations and error management.

These capabilities overlap technologically but are not synonyms. Calling all of them a quantum internet conceals the scientific result.

Whether boson-sampling advantage is durable

Large optical sampling experiments probe distributions believed hard to sample classically. Runtime estimates depend on the best known algorithms, available hardware, required approximation error, device loss, and the verification statistic.

Classical algorithms have improved after major experiments, including methods exploiting photon loss and tensor-network structure. This does not retroactively make the experiments unimportant. It means the defensible claim is dated and task-specific.

What counts as useful non-Gaussianity

A non-Gaussian marginal distribution, a negative Wigner function, a high-fidelity cat state, and a fault-tolerant GKP magic resource answer different questions. Wigner negativity can be a valuable resource witness, but an architecture also needs generation probability, mode quality, stability, and compatibility with its gates and decoder.

“Non-Gaussian” is a classification, not an application benchmark.

Whether more squeezing always helps

At the source, larger squeezing lowers one ideal quadrature variance and raises the conjugate variance. Loss pulls the squeezed variance toward vacuum. Phase noise couples the large anti-squeezed variance back into the measured quadrature. Pump-induced technical noise and detector nonlinearity can also worsen.

An instrument therefore has an operating optimum. A loss-corrected source record need not maximize task performance.

Whether integrated means scalable

Integration reduces path length and alignment count, enables foundry fabrication, and can improve stability. Scaling additionally requires yield, packaging, thermal management, low-loss interfaces, test access, control electronics, detector integration, and repair or redundancy strategies.

A four-mode monolithic circuit and a high-yield million-component system are separated by more than mode count.

Whether postselection preserves the claimed task

Postselection can prepare clean conditional states and reveal fundamental interference. It can also open detection loopholes, hide loss, or change a sampling distribution. Its accept rule must be part of the protocol.

For a communication or computation claim, the discarded trials consume time and resources. For a foundational claim, fair-sampling assumptions may alter what is established.

Whether quantum reservoirs provide advantage

Quantum optical reservoirs can have controllable memory and rich correlations. A quantum advantage requires a matched classical baseline and an end-to-end cost model. Classical preprocessing, repeated measurements, training data, shots, optical energy, feedback electronics, and the digital twin must all be counted.

At this review date, experimental capability is established; general advantage is active and unresolved.

Key Experimental Platforms

Platform	Native resource	Current strength	Main bottleneck	Defensible frontier claim
cavity-coupled quantum dots	triggered single photons and spin–photon states	brightness, indistinguishability, fast repetition	spectral diffusion, device matching, extraction, spin coherence	high-quality photons from selected devices
molecules and colour centres	narrow optical transitions and local memories	tunability, long coherence, nanophotonic coupling	yield, spectral stability, cryogenics, collection	independent-emitter interference and node primitives
bulk SPDC and optical parametric oscillators	pairs, squeezing, Gaussian entanglement	mature mode control and low technical noise	footprint, stabilization, probabilistic conditioning	calibrated high-quality source states
integrated SFWM and SPDC	pairs and squeezed modes on chip	multiplexing, foundry compatibility, compact filters	pump rejection, loss, thermal drift, coupling	co-integrated generation and processing
silicon and silicon-nitride circuits	passive circuits, sources, detectors, GKP synthesis	dense fabrication and telecom compatibility	switching, interfaces, cryogenic control, yield	scalable component technology, not yet a full computer
lithium-niobate photonics	fast modulation, nonlinear conversion, squeezing	electro-optic bandwidth and low-loss routing	source–detector co-integration and fabrication spread	strong control and transduction primitives
atomic ensembles and trapped atoms or ions	memories and matter–photon entanglement	long coherence, high-fidelity local control	collection, repetition rate, deployed-link complexity	elementary memory links and repeater primitives
rare-earth and defect nanophotonics	multiplexed solid-state memories	many emitters in compact cavities	inhomogeneity, optical stability, local-gate scaling	multi-emitter node functionality
fibre and free-space links	long-distance transmission	mature telecom hardware and deployed reach	exponential loss, phase and polarization drift	direct transmission or elementary links
superconducting detectors	efficient photon and number resolution	low dark counts, timing, high efficiency	cryogenics, recovery time, array readout	high-quality terminal measurement
balanced homodyne receivers	quadrature measurement	efficiency, bandwidth, room-temperature electronics	local-oscillator mode matching and calibration	deterministic Gaussian readout
microwave–optical transducers	hybrid-network conversion	connection to superconducting processors	efficiency, added noise, pump heating	coherent conversion; quantum-channel claims need stronger tests

Platform comparisons should hold the task fixed. A detector-free room-temperature Gaussian receiver and a cryogenic photon-number-resolving receiver solve different measurement problems.

Key Theoretical Tools

Mode decompositions

Schmidt decompositions identify the spectral-temporal modes of pair sources. Singular-value decompositions identify which input and output modes an interferometer couples. A scalar “bandwidth” does not replace either.

Correlation functions

$g^{(1)}$ and $g^{(2)}$ connect coherence and count statistics to an operational detector model. Higher-order correlations diagnose multiphoton structure and many-mode sampling devices.

Input–output and scattering theory

Input–output equations connect intracavity dynamics to propagating fields. Scattering matrices describe passive circuits and few-photon interactions. Both make the reference plane explicit.

Gaussian phase-space methods

Covariance matrices, symplectic spectra, characteristic functions, and Gaussian channels scale efficiently with mode number. They are indispensable for squeezed networks, loss propagation, and continuous-variable entanglement.

Quasiprobability reconstruction

Homodyne tomography, pattern functions, maximum-likelihood reconstruction, and Bayesian methods estimate optical states. Physicality constraints and regularization can bias reconstructions, so uncertainty belongs on derived negativity and fidelity.

Quantum trajectories

Photon counting and homodyne monitoring condition an open system on a measurement record. Trajectory theory separates no-click evolution, jumps, diffusive backaction, and feedback.

Channel theory

Pure-loss, thermal-loss, erasure, dephasing, and conversion channels connect component measurements to communication and fault-tolerance questions. Channel composition exposes how a benign loss at each stage becomes a severe end-to-end erasure probability.

Resource theories

Nonclassicality, non-Gaussianity, Wigner negativity, entanglement, coherence, and magic are distinct resources. Monotones help determine whether a declared free operation can create the claimed resource.

Bosonic error correction

GKP, cat, binomial, and related codes encode logical systems in oscillator modes. Finite energy, loss, dephasing, leakage, syndrome extraction, and decoder performance determine logical behaviour.

Network queuing and scheduling

Repeater rates are stochastic. Markov chains, renewal processes, queueing models, and Monte Carlo simulation capture waiting-time distributions, memory decay, multiplexing, and asynchronous swaps more faithfully than one mean probability.

Complexity and classical simulation

Permanents, hafnians, tensor networks, phase-space sampling, matrix-product states, and approximate samplers define the classical comparison for photonic processors. The relevant classical algorithm changes with loss, connectivity, photon number, and requested error.

Statistical model criticism

Likelihood ratios, held-out observables, calibration uncertainty, drift tests, adversarial spoofers, and preregistered acceptance rules distinguish state validation from agreement with a convenient model.

End-to-end optimization

Architectural studies combine optical loss, source statistics, detector behaviour, switching, feed-forward, code thresholds, clock rate, and fabrication yield. Optimizing one component in isolation can move the bottleneck elsewhere.

Representative Papers and Reviews

The list is selective. Stable textbook derivations remain in the canonical pages linked above.

Foundations and field-wide reviews

• R. J. Glauber, “The Quantum Theory of Optical Coherence,” Physical Review 130, 2529–2539 (1963). Operational correlation hierarchy.
• C. K. Hong, Z. Y. Ou, and L. Mandel, “Measurement of Subpicosecond Time Intervals between Two Photons by Interference,” Physical Review Letters 59, 2044–2046 (1987). Canonical two-photon interference.
• S. L. Braunstein and P. van Loock, “Quantum Information with Continuous Variables,” Reviews of Modern Physics 77, 513–577 (2005). Comprehensive continuous-variable review.
• C. Weedbrook et al., “Gaussian Quantum Information,” Reviews of Modern Physics 84, 621–669 (2012). Gaussian states, channels, protocols, and security.
• P. Lodahl, S. Mahmoodian, and S. Stobbe, “Interfacing Single Photons and Single Quantum Dots with Photonic Nanostructures,” Reviews of Modern Physics 87, 347–400 (2015). Authoritative emitter–nanophotonic interface review.
• S. Wehner, D. Elkouss, and R. Hanson, “Quantum Internet: A Vision for the Road Ahead,” Science 362, eaam9288 (2018). Network capability stack and research roadmap.

Sources, squeezing, and integrated photonics

• Y.-M. He et al., “On-Demand Semiconductor Single-Photon Source with Near-Unity Indistinguishability,” Nature Nanotechnology 8, 213–217 (2013). Resonantly driven quantum-dot source.
• N. Somaschi et al., “Near-Optimal Single-Photon Sources in the Solid State,” Nature Photonics 10, 340–345 (2016). Brightness, purity, and indistinguishability in a micropillar source.
• H. Vahlbruch et al., “Detection of 15 dB Squeezed States of Light,” Physical Review Letters 117, 110801 (2016). Calibrated laboratory squeezing.
• D. Ganapathy et al., “Broadband Quantum Enhancement of the LIGO Detectors with Frequency-Dependent Squeezing,” Physical Review X 13, 041021 (2023). Full-scale frequency-dependent squeezing.
• PsiQuantum Team, “A Manufacturable Platform for Photonic Quantum Computing,” Nature 641, 876–883 (2025). Integrated component stack with explicit conditional benchmarks.
• T. Huang et al., “On-Chip Quantum Interference of Indistinguishable Single Photons from Integrated Independent Molecules,” Nature Nanotechnology 20, 1748–1756 (2025). Tunable independent molecular emitters.
• P. Zarkos et al., “Scalable Feedback Stabilization of Quantum Light Sources on a CMOS Chip,” Nature Electronics 8, 620–630 (2025). Integrated source stabilization.

Continuous variables and non-Gaussian resources

• A. Furusawa et al., “Unconditional Quantum Teleportation,” Science 282, 706–709 (1998). Landmark continuous-variable teleportation.
• S. D. Bartlett et al., “Efficient Classical Simulation of Continuous-Variable Quantum Information Processes,” Physical Review Letters 88, 097904 (2002). Gaussian classical-simulation boundary.
• N. C. Menicucci et al., “Universal Quantum Computation with Continuous-Variable Cluster States,” Physical Review Letters 97, 110501 (2006). Cluster-state architecture.
• M. V. Larsen et al., “Deterministic Generation of a Two-Dimensional Cluster State,” Science 366, 369–372 (2019). More than 30,000 temporal modes in a Gaussian cluster.
• D. Gottesman, A. Kitaev, and J. Preskill, “Encoding a Qubit in an Oscillator,” Physical Review A 64, 012310 (2001). Original grid-state code.
• A. Ourjoumtsev et al., “Generating Optical Schrödinger Kittens for Quantum Information Processing,” Science 312, 83–86 (2006). Photon-conditioned non-Gaussian optical states.
• L. S. Madsen et al., “Quantum Computational Advantage with a Programmable Photonic Processor,” Nature 606, 75–81 (2022). Borealis Gaussian boson sampling.
• C. Oh et al., “Classical Algorithm for Simulating Experimental Gaussian Boson Sampling,” Nature Physics 20, 1461–1468 (2024). Loss-aware classical baseline.
• M. V. Larsen et al., “Integrated Photonic Source of Gottesman–Kitaev–Preskill Qubits,” Nature 642, 587–591 (2025). Integrated optical grid-state generation.
• X. Jia et al., “Monolithic Integration of Continuous-Variable Cluster-State Generation, Manipulation and Measurement,” Nature Photonics 20, 428–436 (2026). Four-mode on-chip cluster-state experiment.
• I. Paparelle et al., “Experimental Memory Control in Continuous-Variable Optical Quantum Reservoir Computing,” Nature Photonics 20, 413–420 (2026). Multiplexed squeezed-state reservoir with feedback.
• Y.-H. Deng et al., “Gaussian Boson Sampling with 1,024 Squeezed States in 8,176 Modes,” Nature 653, 687–692 (2026). Jiuzhang 4.0 scale and current complexity claim.

Quantum networks

• H.-J. Briegel et al., “Quantum Repeaters: The Role of Imperfect Local Operations in Quantum Communication,” Physical Review Letters 81, 5932–5935 (1998). Repeater architecture.
• C. M. Knaut et al., “Entanglement of Nanophotonic Quantum Memory Nodes in a Telecom Network,” Nature 629, 573–578 (2024). Two-node solid-state memory network.
• A. Ruskuc et al., “Multiplexed Entanglement of Multi-Emitter Quantum Network Nodes,” Nature 639, 54–59 (2025). Multi-emitter rare-earth nodes.
• W.-Z. Liu et al., “Long-Lived Remote Ion–Ion Entanglement for Scalable Quantum Repeaters,” Nature 652, 51–57 (2026). Memory lifetime beyond mean link-establishment time.
• Y. Zheng et al., “Large-Scale Quantum Communication Networks with Integrated Photonics,” Nature 651, 68–75 (2026). Twenty-client integrated twin-field QKD network.

Common Misconceptions

A low value of g²(0) means a perfect source

It constrains equal-time multiphoton statistics under a declared detector model. It does not determine extraction efficiency, indistinguishability, spectral purity, brightness, or stability.

A heralded source is deterministic

Heralding tells the experiment when a probabilistic event probably occurred. Deterministic delivery additionally requires high generation, herald, routing, and output efficiencies.

High Hong–Ou–Mandel visibility proves complete identity

The visibility probes overlap in the modes resolved by the experiment and is affected by multiphoton events and corrections. Hidden distinguishability outside the receiver bandwidth may not be tested.

Fifteen decibels at the source means fifteen decibels in an instrument

Propagation loss, imperfect visibility, detector inefficiency, and phase noise reduce observable squeezing. The application-level value is measured at the operating reference plane.

Gaussian entanglement alone gives universal computation

Gaussian states, Gaussian transformations, and Gaussian measurements remain within a restricted class. Universal processing requires a non-Gaussian ingredient and a fault-tolerant strategy for finite squeezing and loss.

Wigner negativity proves fault tolerance

Negativity establishes a strong nonclassical feature. Fault tolerance requires a code, noise model, syndrome procedure, decoder, and decreasing logical error with increasing resources.

More optical modes mean more useful computation

Modes can be empty, inaccessible, weakly connected, highly lossy, or classically tractable. Useful scale includes occupation, programmability, connectivity, fidelity, detection, and task complexity.

A quantum network is any setup with several optical terminals

The capability must be named: key distribution, entanglement distribution, state teleportation, memory networking, repeater operation, distributed sensing, or distributed computation.

Integrated photonics eliminates alignment and calibration

It replaces some mechanical alignment with fabrication, packaging, thermal tuning, electronic calibration, and process-control problems. The burden changes form.

A conditional fidelity can be multiplied by a rate later

The conditional state and the success event can be correlated. Background, detector saturation, source brightness, and drift may change fidelity as the rate changes. Joint operating-point data are preferable.

A sampling advantage settles universal quantum advantage

It addresses a specified sampling problem relative to specified classical algorithms and accuracy. It does not establish universal computation or advantage on unrelated applications.

Postselection is merely a data-analysis detail

Postselection defines the implemented channel and its resource cost. In foundational tests it can also change which loopholes are closed.

Exercises

Exercise 1: Audit a single-photon source claim

A pulsed source reports $g^{(2)}(0)=0.008$, a first-lens brightness of $0.60$ photons per pulse, fibre coupling of $0.75$, detector efficiency of $0.90$, and Hong–Ou–Mandel visibility $0.96$ after background correction. The laser repetition rate is $80\ \mathrm{MHz}$.

1. Estimate the detected one-photon rate using the stated factors.
2. Which reported number diagnoses multiphoton contamination?
3. Which additional values are needed before calling the source mutually indistinguishable with ten independently fabricated sources?

Solution

The idealized detected rate is

$$\begin{aligned} R_{\mathrm{det}} &= (80\times10^6\ \mathrm{s}^{-1}) (0.60)(0.75)(0.90) \\ &= 3.24\times10^7\ \mathrm{s}^{-1}. \end{aligned}$$

This estimate assumes that brightness, coupling, and detector efficiency refer to compatible modes and operating conditions and that saturation and dead time are negligible.

The value $g^{(2)}(0)$ diagnoses equal-time multiphoton correlations under the source and detector model. It does not alone give the one-photon probability.

The two-source Hong–Ou–Mandel result does not establish mutual indistinguishability across ten devices. One also needs pairwise or network-relevant mode-overlap data, tuning ranges, spectral-diffusion and drift statistics, brightness and purity at the same operating point, polarization and temporal-mode control, raw and corrected visibilities, and the yield of devices meeting the specification.

Exercise 2: Multipair contamination

An ideal two-mode squeezed pair source has mean pair number $\mu=0.05$ per pulse.

1. Find $q$ in the geometric distribution used above.
2. Find the probability of at least one pair.
3. Find the multipair fraction conditioned on at least one pair.

Solution

From $\mu=q/(1-q)$,

$$q = \frac{\mu}{1+\mu} = \frac{0.05}{1.05} \approx 0.047619.$$

Because $P_0=1-q$,

$$P(n\ge1) = q \approx 0.047619.$$

For a geometric pair distribution,

$$P(n\ge2\mid n\ge1) = q \approx 4.76\%.$$

The equality of the last two numerical values is a property of this parameterization and distribution. Finite herald efficiency and number-resolution change the experimentally conditioned fraction.

Exercise 3: Multiplexing with switch loss

Thirty-two independent source bins each succeed with probability $p=0.05$. A binary routing tree has five switch stages, each with transmission $\tau=0.98$.

1. Find the probability that at least one source succeeds.
2. Assuming one selected photon crosses all five stages, estimate the useful delivery probability.
3. Name two costs absent from this estimate.

Solution

The ideal success probability is

$$P_{\ge1} = 1-(1-0.05)^{32} = 1-0.95^{32} \approx 0.8063.$$

The switching transmission is

$$\eta_{\mathrm{switch}} = 0.98^5 \approx 0.9039.$$

The simplified delivered probability is

$$P_{\mathrm{del}} \approx (0.8063)(0.9039) \approx 0.7288.$$

Missing costs include source-to-switch coupling, delay-line loss, feed-forward latency, detector efficiency in the herald arm, simultaneous success collisions, mode mismatch among bins, switch crosstalk, and the clock cycles consumed while routing.

Exercise 4: Squeezing after loss

A source produces $15\ \mathrm{dB}$ of squeezing relative to vacuum. The total transmission to the measurement plane is $\eta=0.90$ and all loss ports inject vacuum.

1. Find the input variance ratio $V_{\mathrm{in}}/V_{\mathrm{vac}}$.
2. Find the output ratio.
3. Express the observable squeezing in decibels.

Solution

The source ratio is

$$\frac{V_{\mathrm{in}}}{V_{\mathrm{vac}}} = 10^{-15/10} = 10^{-1.5} \approx 0.03162.$$

Pure loss gives

$$\begin{aligned} \frac{V_{\mathrm{out}}}{V_{\mathrm{vac}}} &= (0.90)(0.03162)+0.10 \\ &\approx 0.12846. \end{aligned}$$

Therefore

$$S_{\mathrm{out}} = -10\log_{10}(0.12846) \approx 8.91\ \mathrm{dB}.$$

Ten percent loss reduces a $15\ \mathrm{dB}$ source to less than $9\ \mathrm{dB}$ at the receiver. Phase noise would reduce it further by mixing in anti-squeezed fluctuations.

Exercise 5: Raw and corrected interference visibility

A Hong–Ou–Mandel measurement gives $C_{\mathrm{dist}}=1000$ coincidences outside the dip and $C_{\mathrm{ind}}=30$ at the dip. An independently measured accidental background contributes 20 coincidences to each setting.

1. Find the raw visibility.
2. Find the background-corrected visibility.
3. Which value should appear in an end-to-end benchmark?

Solution

The raw value is

$$V_{\mathrm{raw}} = \frac{1000-30}{1000} = 0.970.$$

After subtracting 20 from each count,

$$\begin{aligned} V_{\mathrm{corr}} &= \frac{980-10}{980} \\ &\approx 0.9898. \end{aligned}$$

Both should be reported with the background model and uncertainty. The raw value describes the observed system output. The corrected value estimates intrinsic overlap under the assumption that the independently estimated background is additive and correctly modelled. A system benchmark should not silently replace the raw operating performance with the corrected value.

Exercise 6: Direct fibre loss and a midpoint protocol

Telecom fibre has attenuation $\alpha=0.2\ \mathrm{dB/km}$. Two memories are separated by $100\ \mathrm{km}$.

1. Find the end-to-end channel transmission.
2. In a symmetric midpoint protocol, each memory emits one photon toward a central station. Find each half-link transmission and their product.
3. Why is the product not yet the entanglement-generation probability?

Solution

The full-link transmission is

$$\eta_{100} = 10^{-(0.2)(100)/10} = 10^{-2} = 0.01.$$

Each half is $50\ \mathrm{km}$:

$$\eta_{50} = 10^{-(0.2)(50)/10} = 10^{-1} = 0.10.$$

If the protocol requires both emitted photons to arrive, the channel factor is

$$\eta_{50}^2 = 0.01.$$

The full success probability also includes emission into the desired mode, coupling, frequency conversion, detector efficiency, interference visibility, Bell-measurement success, temporal acceptance, and memory preparation. Some single-photon heralding protocols have a different transmission scaling but add phase-stability and false-herald constraints.

Exercise 7: Gaussian closure

A processor begins with squeezed-vacuum states, applies only beam splitters, phase shifts, displacements, and additional squeezers, transmits modes through pure-loss channels, and measures every output by homodyne detection.

1. Is the unconditional state before measurement Gaussian?
2. Are the homodyne outcome statistics described by Gaussian conditional rules?
3. Name one change that can introduce a non-Gaussian resource.

Solution

Yes. Squeezed vacuum is Gaussian; Gaussian unitaries and pure-loss Gaussian channels preserve Gaussianity. Homodyne detection is a Gaussian measurement, and conditioning on its continuous outcome updates means and covariances within the Gaussian formalism.

Possible non-Gaussian changes include photon-number-resolving measurement, photon subtraction or addition, injection of a cat or GKP state, or a sufficiently nonlinear matter-mediated interaction. Merely increasing the number of Gaussian modes does not leave the Gaussian class.

Exercise 8: Classify four frontier claims

Classify each claim as established, active, conjectural, speculative, or controversial, and state the missing evidence.

1. A chip generates and verifies a four-mode continuous-variable cluster state.
2. The same fabrication process will support a fault-tolerant million-mode computer.
3. A 20-client twin-field QKD test is a 20-node universal quantum internet.
4. A sampling processor outpaces the best classical algorithm evaluated in its paper for a declared distribution and accuracy.

Solution

1. Established for the measured chip, modes, observables, and uncertainty if the entanglement witness and calibration support the claim.
2. Conjectural as an engineering extrapolation. It needs yield, loss, control, non-Gaussian resources, error correction, decoder, and logical scaling evidence.
3. The experimental QKD capability can be established, but the universal quantum-internet interpretation is false or at best speculative. The experiment does not demonstrate arbitrary state transfer, stored entanglement, swapping, or distributed quantum gates among 20 processors.
4. Established relative to the tested classical baselines and stated assumptions at publication. A broader or permanent computational advantage is active and may be controversial if experts dispute approximation error, validation, or the best classical comparison.

The scope clause is part of every classification.

Connections to Other Volumes

Composite systems and entanglement

Bell States, Two-Mode Entanglement, and Entanglement in Quantum Optics provide the state and witness language used in optical networks.

Measurement and open systems

Photon Counting and Homodyne Detection connect optical records to conditioned dynamics. Channel pages supply the right language for loss, added noise, and conversion.

Dynamics and phase space

Gaussian States and Wigner Functions develops efficient phase-space evolution. Non-Gaussian optical resources mark where moment closure and positive Gaussian descriptions cease to be enough.

Experimental foundations

Loophole-Free Bell Tests show why detection efficiency, event-ready heralding, locality, and postselection are not optional details in foundational optical claims.

Canonical systems

Coherent States and A First Encounter with Squeezed States provide the oscillator foundation. The frontier begins when those states are embedded in multimode, lossy, measured systems.

Atomic and cavity interfaces

Cavity QED and Rydberg Array Frontiers connect flying optical modes to stationary emitters and processors. Cavity and Circuit QED Frontiers owns ultrastrong, multimode, waveguide, superconducting-circuit, hybrid-link, and bosonic-memory questions.

What Changed This Year

Long-lived ion–ion entanglement crossed a repeater-relevant boundary

In February 2026, remote trapped-ion memories connected through $10\ \mathrm{km}$ of fibre retained entanglement beyond the mean establishment time. This moves a key repeater criterion from proposal to experiment. It does not yet demonstrate a multi-segment repeater or a finite-rate end-to-end advantage over direct transmission.

Integrated QKD expanded to twenty client chips

Also in February 2026, a silicon-nitride microcomb server and 20 indium-phosphide client transmitters implemented pairwise twin-field QKD through ten wavelength channels over $370\ \mathrm{km}$ spools. The advance is integrated, reproducible, many-client secure-key networking. It should not be relabelled as stored-entanglement networking.

Continuous-variable generation, gates, and readout were monolithically combined

In March 2026, a single integrated circuit generated squeezed light, applied single-mode and two-mode Gaussian operations, supplied local oscillators, and performed balanced homodyne detection, verifying four-mode cluster entanglement. The immediate frontier is scaling this closed loop while retaining loss, phase, and calibration performance and adding non-Gaussian resources.

A squeezed optical reservoir gained controllable memory

A March 2026 experiment combined spectral and temporal multiplexing, multimode squeezing, mode-selective homodyne detection, electro-optic feedback, and a calibrated digital twin. It established controllable fading memory and temporal-task capability. Quantum computational advantage remains an open comparison, not a consequence of using squeezed states.

Gaussian boson sampling increased sharply in optical scale

Jiuzhang 4.0 reported 1,024 squeezed sources, 8,176 modes, $51\%$ overall system efficiency, and events with up to 3,050 detected photons. The result substantially raises the experimental scale and updates the classical simulation contest. It remains a programmable sampling processor, not a universal fault-tolerant machine.

Optical GKP generation became a baseline rather than a proposal

The 2025 integrated GKP source is now part of the field’s experimental baseline. The correct 2026 question is no longer whether propagating optical grid structure can be generated on chip. It is whether loss, generation probability, refinement, switching, and logical processing can be combined so that an encoded error decreases under a complete protocol.

No single 2026 result resolves the frontier. Together they shift emphasis from isolated optical primitives toward integrated source–circuit–detector loops, memory-aware networks, and explicit resource accounting.