Three-Photon Distillation Protocol

• The protocol employs a heralded linear-optical method using three imperfect photons to reduce the indistinguishability error from ε to approximately ε/3.
• It utilizes a 3×3 Fourier network and postselection to enhance Hong–Ou–Mandel visibility, demonstrating key improvements for photonic quantum computing.
• Experimental implementations in integrated photonic chips confirm scalable error reduction and validate the protocol as a crucial primitive for near-term quantum technologies.

A three-photon distillation protocol is a heralded linear-optical procedure that uses three imperfect photons to increase photonic indistinguishability by interference and postselection. In the formulation that has become standard for photonic quantum computing, three input states $\rho(\epsilon)^{\otimes 3}$ are interfered in a $3\times 3$ Fourier network and two output modes are measured, producing, on success, a single photon whose indistinguishability error is reduced from $\epsilon$ to $\epsilon/3+\mathcal{O}(\epsilon^2)$ with success probability $1/3$ in the low-error regime. A distinct but closely related formulation uses three photons to herald a distilled two-photon pair with increased Hong–Ou–Mandel visibility, and in that setting the full three-photon Gram matrix, including the triad phase, is operationally relevant (Somhorst et al., 2024, Hoch et al., 2 Sep 2025).

1. Definition, scope, and main variants

The term denotes a family of protocols rather than a single circuit. The earliest constructive three-photon example used three imperfect photons, standard linear optics, and post-selection to output one heralded photon with higher indistinguishability, with asymptotic resource scaling $O((\epsilon/\epsilon')^2)$ under repeated use (Marshall, 2022). Subsequent work identified that construction as the $n=3$ member of a systematic family based on multiphoton Fourier interference, in which an $N$-photon protocol achieves $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$ and has a heralding probability approaching $1/4$ as $3\times 3$0 (Somhorst et al., 2024). A later line of work treated the minimal non-trivial case of three photons with arbitrary partial distinguishability and optimized the output HOM visibility under a fixed heralding pattern, proving that three spatial modes suffice for optimality (Hoch et al., 2 Sep 2025).

The literature therefore uses the same label for two non-equivalent tasks. One task is single-photon purification in the orthogonal bad-bit model; the other is scenario-dependent optimization of a heralded photon pair’s HOM visibility. A plausible implication is that “three-photon distillation protocol” should always be read together with its figure of merit.

Formulation	Three-photon input structure	Primary figure of merit
Fourier single-photon distillation	Three imperfect photons, one per input mode	Output error $3\times 3$1
Optimal pair distillation	Three photons with Gram matrix $3\times 3$2	Distilled visibility $3\times 3$3
Experimental fault-tolerant benchmark	Three distillation photons plus a reference photon for characterization	$3\times 3$4, $3\times 3$5

This dual usage is historically important. The 2024 Fourier-based framework unified earlier small circuits with a scalable family (Somhorst et al., 2024), whereas the 2025 optimization study showed that for generic three-photon distinguishability structures the older symmetric circuit is not generally optimal (Hoch et al., 2 Sep 2025).

2. State model and the meaning of indistinguishability distillation

In the Fourier-based single-photon framework, an imperfect source is modeled by the orthogonal bad-bit model,

$3\times 3$6

where $3\times 3$7 is a pure fully indistinguishable photon, $3\times 3$8 is an orthogonal error mode treated as fully distinguishable, and $3\times 3$9 is the partial distinguishability error. For $\epsilon$0 such sources the joint state is $\epsilon$1, and for sufficiently small $\epsilon$2 it is expanded to first order as

$\epsilon$3

The probability of higher-order errors is

$\epsilon$4

and the low-error approximation is valid when $\epsilon$5 (Somhorst et al., 2024).

Photon distillation is then defined operationally. A linear-optical circuit implements a unitary $\epsilon$6 on the external modes, the output state is partially measured in $\epsilon$7 modes, and the conditional state of the remaining mode is normalized. If the herald pattern has total detected photon number $\epsilon$8, the remaining mode is projected onto a single-photon state with new error parameter $\epsilon$9. In this usage, “photon distillation” means any choice of $\epsilon/3+\mathcal{O}(\epsilon^2)$0 and herald measurement such that $\epsilon/3+\mathcal{O}(\epsilon^2)$1 (Somhorst et al., 2024).

A related 2022 formulation uses a random-phase/error model in which the indistinguishability is

$\epsilon/3+\mathcal{O}(\epsilon^2)$2

so increasing indistinguishability is equivalent to reducing the error weight in orthogonal internal modes. This suggests that the exact operational proxy depends on the chosen source model, but the common structure is always interference plus postselection on multiphoton events (Marshall, 2022).

3. Three-photon Fourier distillation to a single output photon

For $\epsilon/3+\mathcal{O}(\epsilon^2)$3, the general Fourier construction becomes a three-photon protocol built from the $\epsilon/3+\mathcal{O}(\epsilon^2)$4 discrete Fourier transform

$\epsilon/3+\mathcal{O}(\epsilon^2)$5

Three imperfect photons $\epsilon/3+\mathcal{O}(\epsilon^2)$6 are injected one per mode, the interferometer implements $\epsilon/3+\mathcal{O}(\epsilon^2)$7, modes $\epsilon/3+\mathcal{O}(\epsilon^2)$8 and $\epsilon/3+\mathcal{O}(\epsilon^2)$9 are measured, and mode $1/3$0 is retained as the distilled output. The protocol exploits the zero-transmission law: for fully indistinguishable photons in a Fourier interferometer, an output pattern with ordered mode-assignment list $1/3$1 is forbidden whenever $1/3$2. Valid herald patterns are chosen from the allowed set, and for $1/3$3 the canonical collision-free pattern is $1/3$4, corresponding to one photon in each output mode (Somhorst et al., 2024).

In the low-error regime, the general Fourier analysis gives

$1/3$5

so for three photons

$1/3$6

The total heralding probability is

$1/3$7

and the average photon cost per distilled photon is therefore

$1/3$8

The general heralding probability for $1/3$9 photons tends to $O((\epsilon/\epsilon')^2)$0 as $O((\epsilon/\epsilon')^2)$1, but the three-photon case remains the experimentally simplest nontrivial instance (Somhorst et al., 2024).

The earlier explicit three-photon circuit analysis reached the same asymptotic conclusion through a concrete three-beamsplitter network with two 50:50 beam splitters and one asymmetric beam splitter of angle $O((\epsilon/\epsilon')^2)$2. In that formulation, heralding corresponds to detecting one photon in each of the two measured rails, and the output error obeys

$O((\epsilon/\epsilon')^2)$3

$O((\epsilon/\epsilon')^2)$4

while the post-selection success probability satisfies

$O((\epsilon/\epsilon')^2)$5

These bounds recover the same first-order map $O((\epsilon/\epsilon')^2)$6 and $O((\epsilon/\epsilon')^2)$7 (Marshall, 2022).

The 2024 Fourier treatment also establishes the structural significance of this small circuit. The earlier three-photon protocol is not an isolated numerically found gadget but exactly the $O((\epsilon/\epsilon')^2)$8 case of an $O((\epsilon/\epsilon')^2)$9-photon Fourier family with error-reduction factor $n=3$0 in one step (Somhorst et al., 2024).

4. Optimal three-photon distillation with arbitrary partial distinguishability

A different three-photon protocol starts from three photons with pure internal states $n=3$1 and encodes distinguishability in the Gram matrix

$n=3$2

where $n=3$3 and

$n=3$4

is the triad phase of the three-photon Bargmann invariant. In this formulation, photon 1 is left untouched, photons 2 and 3 enter a 3-mode interferometer $n=3$5 together with one vacuum mode, and the heralded output is a single distilled photon $n=3$6 that is then interfered with photon 1. The target quantity is the distilled HOM visibility

$n=3$7

and the visibility gain is defined as

$n=3$8

A protocol with $n=3$9 genuinely distills indistinguishability in this sense (Hoch et al., 2 Sep 2025).

The central analytic result is that $N$0 and the success probability $N$1 depend only on the input distinguishability data and two interferometer parameters,

$N$2

For pure inputs, the multi-photon contribution enters through a term proportional to

$N$3

so the sum of the triad phase and the unitary phase controls whether three-photon interference helps or hinders the distillation. The work proves that for any given $N$4 there exists a 3-mode unitary $N$5, and among all unitaries reaching the maximal $N$6 the authors analytically choose the one that maximizes $N$7, still with the minimum number of spatial modes equal to three (Hoch et al., 2 Sep 2025).

This scenario-dependent optimization changes the status of older symmetric circuits. In the symmetric real case $N$8 and $N$9, the optimal parameters are $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$0 and $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$1, yielding

$\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$2

However, when the pairwise visibilities are unequal or $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$3, the same $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$4 is not generally optimal and can even give negative gain. In the symmetric case with $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$5, $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$6 gives no gain, whereas the optimal parameters switch to $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$7 and $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$8, which is equivalent to swapping two rows of $\epsilon'=\epsilon/N+\mathcal{O}(\epsilon^2)$9 and postselecting on a different output mode. This establishes that the triad phase is not a perturbative refinement but a decisive control parameter in generic three-photon distillation (Hoch et al., 2 Sep 2025).

5. Experimental realizations and measured performance

The experimentally simplest realizations use integrated photonics. For the Fourier single-photon protocol, losses primarily reduce the success probability, certain loss patterns can produce erasure errors, and detector inefficiencies reduce the effective heralding probability without fundamentally altering the conditional state as long as “no click” is not misinterpreted as “no photon.” A 3-mode DFT is described as straightforward to implement in silica, silicon nitride, or silicon photonic chips using $1/4$0 beamsplitters and phase shifters, and the minimal depth of the 3-mode DFT limits accumulated loss (Somhorst et al., 2024).

The optimized three-photon protocol was experimentally validated with a demultiplexed quantum-dot source interfaced with a programmable eight-mode laser-written integrated photonic processor. The source was a commercial InGaAs quantum-dot resonant-fluorescence source at $1/4$1, a time-to-spatial bulk demultiplexer driven by an AOM produced three simultaneous photons, and the integrated processor supplied universal control over $1/4$2 unitaries through 28 tunable beam splitters implemented by cascaded couplers and thermo-optic phase shifters. For the distillation protocol only modes 3–6 were actively used, with one beam-splitter cell configured as the HOM test stage. The experiment reported positive gain for almost the entire real symmetric $1/4$3 range, restoration of positive gain for real symmetric $1/4$4 when the optimal $1/4$5 was chosen, significant positive gain for complex triad phases $1/4$6, and gains up to $1/4$7 under realistic conditions including multiphoton noise and imperfect unitaries (Hoch et al., 2 Sep 2025).

A later experiment implemented the scalable, optimal photon-distillation family in the three-photon case and directly addressed fault-tolerant relevance. Using a resonantly driven InGaAs quantum dot at $1/4$8, a temporal-to-spatial demultiplexer, and a 20-mode silicon-nitride integrated quantum photonic processor, the experiment realized a 3-mode distillation unitary with matrix fidelity $1/4$9. The measured input errors were

$3\times 3$00

and after three-photon distillation they became

$3\times 3$01

This corresponds to an indistinguishability-error reduction of about $3\times 3$02 and a total-error reduction of about $3\times 3$03, while the effective unitary-induced indistinguishability error was estimated as $3\times 3$04 (Somhorst et al., 9 Jan 2026).

The 2026 work described this regime as unconditional error reduction, or below-threshold behavior, because the total effective error decreased even after accounting for noise introduced by the distillation gate itself. That result is specific to the experimentally characterized noise budget and should not be conflated with the ideal small-$3\times 3$05 asymptotic factor of $3\times 3$06 (Somhorst et al., 9 Jan 2026).

6. Resource trade-offs, quantum-computing role, and recurrent misconceptions

Within photonic quantum computing, partial distinguishability appears as reduced interference visibility, effective Pauli errors, dephasing, and leakage from the code space during resource-state construction and fusion operations. In the simplified model used in the Fourier analysis, gate error rates scale roughly as $3\times 3$07, so a three-photon distillation step that sends $3\times 3$08 also reduces these gate error rates by the same factor, at the expense of a heralded overhead of nine raw photons per distilled single photon in the ideal $3\times 3$09 protocol (Somhorst et al., 2024).

The intended architectural role is therefore pre-processing, not replacement of quantum error correction. The distillation stage is applied before cluster-state generation or fusion-based resource construction, so that the effective physical error entering QEC is lower. In the 2026 fault-tolerance analysis, $3\times 3$10 and the surface-code threshold was taken as $3\times 3$11; under that model there are low-error, intermediate, and above-QEC-threshold regimes, and the three-photon protocol already extends the usable range of deterministic quantum-dot sources while larger $3\times 3$12 improves asymptotic efficiency further (Somhorst et al., 9 Jan 2026).

Several misconceptions recur in the literature. One is that the three-photon protocol is a unique canonical circuit. In fact, the label covers at least two distinct tasks: single-photon purification in the orthogonal bad-bit/Fourier framework and two-photon visibility optimization for arbitrary Gram matrices (Somhorst et al., 2024, Hoch et al., 2 Sep 2025). Another is that the protocol is merely a three-photon version of Hong–Ou–Mandel filtering. The later optimization work shows that pairwise visibilities alone are insufficient: the triad phase can decide whether the same interferometer yields positive gain, no gain, or negative gain (Hoch et al., 2 Sep 2025). A third is that photon distillation replaces fault tolerance. The scalable literature treats it as a coherent bosonic error-mitigation layer that can raise effective thresholds and reduce code distance, but still in tandem with QEC rather than instead of it (Somhorst et al., 9 Jan 2026).

Taken together, the modern meaning of a three-photon distillation protocol is that three-photon interference is the smallest nontrivial setting in which indistinguishability can be probabilistically improved by heralding, either by producing one better single photon with $3\times 3$13 and $3\times 3$14, or by optimally reshaping a three-photon distinguishability structure so as to maximize the HOM visibility of a distilled pair. In both senses, the protocol has become a standard primitive for near-term integrated photonics and a conceptual prototype for larger multiphoton distillation schemes (Marshall, 2022, Somhorst et al., 2024).