WorksPhysics and Astronomy
Pure dephasing destroys two-photon amplitude, not nonlinear phase
Abstract
A two-level emitter chirally coupled to a waveguide imprints a conditional phase on a two-photon pulse, a key component assumed in quantum photonic neural network architectures, but whether it survives the pure dephasing of a solid-state emitter is unknown, since no two-photon scattering treatment includes it. We derive the dephased two-photon amplitude in closed form and validate it against an independent Fock-state master equation. The conditional phase is pinned at exactly at every dephasing rate examined: dephasing destroys the amplitude carrying the nonlinear phase rather than rotating it. The one- and two-photon channels are not attenuated equally, so dephasing does not simply scale the node down but reweights it between photon-number sectors—an error no phase setting can train away. The standard factorization of two-photon correlators, exact for arbitrary loss, fails at first order in , discarding – of pairs over –.
A two-level emitter chirally coupled to a waveguide is the simplest optical nonlinearity that acts at the level of single photons [1, 2, 3, 4, 5], and in the ideal limit it realizes a full conditional phase between one- and two-photon inputs—the ideal Kerr response, and twice the ceiling of a bidirectional geometry [6, 7, 8, 9, 10, 11].
This is no longer an idealization in practical photonic systems: Staunstrup et al. [12] have measured the emitter-induced phase interferometrically in a bidirectional waveguide, and a rapidly growing body of architecture work is built on exactly this response: quantum photonic neural networks [13, 14, 15], chiral quantum networks [16] and cluster-state generators [17], the most recent modeling the emitter with its full chiral scattering matrix [15], but without pure dephasing. Continuous-wave chiral nonlinear optics and all-optical control draw on the same interaction [18, 19]. In a bidirectionally coupled waveguide the phase excursion is bounded by and maximal off resonance; the measurement gives , and Ref. [12] identifies chiral coupling as the route to the full, on-resonance , which is the regime treated here. Quantum emitters also serve as reconfigurable phase elements in otherwise linear circuits [20].
The difficulty is that all solid-state emitters dephase. Inverting the measured phase excursion on the two dipoles of Ref. [12] gives total linewidths 1.4 and 3.2 times the radiative rate, i.e., pure dephasing rates of order and for a typical coupling efficiency [21, 22]. These are not perturbative corrections. More generally, –1 is the range that matters in practice: lower rates have been reached in diode-embedded structures where background charge noise is suppressed electrically [23, 24], but such devices are demanding to integrate at the scale a network requires, and is typical of state-of-the-art emitters in waveguides.
Yet the theory available to describe them is incomplete in a specific way. Frequency-domain two-photon scattering theory, the common language of this field [1, 14, 25, 26, 27], has no dephasing implementation. Matrix-product-state and related non-Markovian methods [28, 29, 30, 31, 32, 33, 34, 35] do include it, and Bundgaard-Nielsen et al. [34] recently gave the single-photon results in full. Their analytic treatment stops there, however, and they state explicitly that it “fails here” for two photons, which is the gate-relevant case, since a nonlinear phase gate is defined by the difference between one- and two-photon response rather than by either alone.
In this Letter, we derive the two-photon scattered amplitude of a chirally coupled emitter driven by a Fock pulse, with pure dephasing included exactly, in closed form, validated against a Fock-state master equation solved with no factorization. Three key results follow: (i) the conditional phase is exactly at every dephasing rate examined, to 2, including above the radiative rate where the single-photon phase has flipped to 0; dephasing costs amplitude, not phase; (ii) the one- and two-photon channels do not attenuate together, so dephasing does not simply dim the node’s output—it changes the balance between its one- and two-photon responses, which alters what the node does to any input containing both; (iii) the factorization of two-photon correlators used throughout the scattering literature is exact for arbitrary loss but fails at first order in , and we give the exact replacement.
**Model.—**We take a two-level system, , coupled to a chiral one-dimensional continuum within the rotating-wave and first Markov approximations. The emitter decays into the guided mode at rate and into all other channels at , so and , and we define with the emitter–pulse detuning. Time-domain amplitudes are written in the pulse frame, so the detuning enters through , while frequency-domain quantities such as are referred to .
The input–output relation is . All results below are quoted for the ideal chiral case , so , though Eq. (3) holds for arbitrary [36]. The drive is an -photon Fock pulse, with and a normalized mode, taken as a Gaussian throughout with amplitude width , .
Pure dephasing can be modeled as classical multiplicative noise with and , equivalent on averaging to the Lindblad term . Per noise realization the dynamics are unitary and the state is pure, so every object may be constructed for a pure state and averaged at the end; a superscript denotes a quantity evaluated in one such realization. Writing for the average over realizations, the distinction that organizes everything is whether an observable is a first moment or a second moment , since in general these are not related by .
For a Fock state , so no ratio exists and the phase must be defined interferometrically. The measurable object is the cross-arm correlator , with a field operator for the reference arm that bypasses the emitter [Fig. 1(a)]. Since that arm carries no dephasing noise, the average factorizes, , a first moment: no factorization is needed, and the result is exact under dephasing with the single substitution .
A Fock pulse enters a waveguide containing a single emitter subject to pure dephasing at rate [Fig. 1(a)]; the coupling is chiral, so the emitter radiates only forward and there is no reflected field to account for. Panels (b) and (c) contrast one and two photons under otherwise identical conditions. In both cases the output separates into a coherent part, drawn inverted because it carries the shift, and an incoherent remainder (dashed curve).
the shift, and an incoherent remainder (dashed curve). Increasing transfers weight from the first to the second while leaving the sign of the coherent part untouched; this is the central result, made quantitative below. The two rows differ in how fast that transfer happens, which is the second result: the coherent two-photon amplitude survives to larger than the one-photon amplitude does. Photon number is conserved throughout; what is lost is phase coherence with the drive, not photons.
Single-photon response.—For the chiral emitter system, the coherent single-photon transmission is
reproducing the result of [34] and, at , the ideal chiral result , . On resonance is real, so can only take the values or : it is a sign, not an angle, and a sign cannot rotate. It is for all , undefined at where vanishes, and above. We stress that is the coherent amplitude, a first moment, and its vanishing is not a loss of photons: the geometry is chiral so nothing is reflected, pure dephasing scatters nothing out of the waveguide, and the transmitted flux obeys exactly at every , a statement about the signal arm alone, with no interferometer present. At the input and the antiphase re-radiated field cancel exactly, so the photons still emerge but with their phase scrambled, in the incoherent component; a photon counter registers them, an interferometer does not.
Physically, one has with the coherently re-radiated field. Dephasing is phase diffusion: it damps the emitter coherence isotropically, adding to the decay rate without contributing anything imaginary, so it shrinks without rotating it. The is an overshoot; at the emitter radiates against an input of and the sum flips sign. While the antiphase amplitude still exceeds the input the sum stays flipped; at the two cancel exactly and there is no coherent field left to carry a phase.
The distinction is between first and second moments, since Eq. (1) is a first moment and can cancel, whereas photon number is a second moment and cannot. The re-emitted light appears as the incoherent Lorentzian of [34]. Figure 2 shows the resulting collapse.
Figure 2. Collapse of the coherent amplitude: single-photon (a)–(c), two-photon (d). Throughout, lines are the closed-form results and symbols the independent Fock-state master equation (Fock ME) of the End Matter. (a) Transmission amplitude and (b) phase from Eq. (1), for , with circles the Fock ME (agreement better than ); the dashed curve is the single above-threshold case , where has flipped to 0. exactly for every . The slope at resonance, , diverges at where vanishes—the is then carried by an amplitude of zero. (c) against . (d) Coherent fraction of the two-photon amplitude, ; the closed form Eq. (3) against the Fock ME, which agree to across the range.
Closed-form two-photon amplitude.—Since is linear in the scattered amplitude, the two-photon phase needs only single amplitudes from the form , with the vacuum already in the bra. For these the intervals over which the emitter accumulates a random phase are disjoint, so the noise average can be taken exactly, with no factorization.
The two-photon output amplitude is , the probability amplitude for detecting one transmitted photon at and a second at . Substituting the input–output relation for each generates four contributions, labelled by how many of the two photons have interacted with the emitter,
where is the amplitude for both photons to pass through untouched, and carry one emitter operator each (one photon scattered, one not) and is the term in which both are scattered. Solving the hierarchy for each and collecting terms gives
where , is the transmitted single-photon amplitude, and is the emitter coherence, which carries the dephasing through . Equation (3) is the central analytic result: an -matrix-like transfer amplitude for a dephasing node, which can be chained to describe several nodes in sequence, valid for arbitrary and arbitrary pulse shape, including entangled and spectrally shaped inputs [37].
Validation is threefold, and detailed in the End Matter. First, against a Fock-state master equation solved with no factorization, the relative error is at rising only to at . Second, against frequency-domain scattering theory at , the worst disagreement over thirteen points is . Third, at the closed form collapses to a single quadrature that needs no time or frequency grid, giving against from scattering theory. The nonlinear term is not a correction: it reaches 2.0–3.4 times the linear one, which is why the two-photon response is nothing like .
Conditional phase
A -phase gate is characterized by the phase of the two-photon amplitude relative to independent single-photon scattering,
with . The entire conditional phase comes from the bound-state term. On resonance with real everything in Eq. (4) is real, so exactly [38], and we find at every dephasing rate examined, to 2, for our chosen input pulse [Figure 3(a)] [39]. The point is excluded there only because the linear reference vanishes, making the ratio singular even though is finite. The value requires the bound-state term to exceed . For a Gaussian pulse at a pulse width of this holds out to and fails only once the emitter has decoupled, where the node is in any case nearly transparent.
Figure 3. The conditional phase across frequency and dephasing, and the unequal attenuation of the two channels. (a) Conditional phase along the two-photon diagonal, , unwrapped through , for (solid), (orange dashed), , (dotted) and (dashed, above threshold). At it is exactly or for any dephasing rate [38] and equals at every rate examined; it is antisymmetric about , so a symmetric resonant pulse returns on averaging. Its width is set by the linear reference , which vanishes at as ; the sweep narrows to , is undefined at , and reverses above threshold. (b) Resonant two-photon transmission at (squares, Fock-state master equation). (c) Resonant coherent transmission of the one-photon channel, , and the two-photon channel, , each normalized to its value, against the independent-scattering expectation . (d) Ratio of the two-photon transmission to that expectation, normalized as in (c), the factor by which the two-photon channel outlives two independent single-photon scatterings.
The penalty appears entirely in the amplitude. The resonant two-photon transmission falls monotonically, for [Figure 3(b)], and the coherent fraction falls from 1 to 0.19 over [Figure 2(d)] while the photon number stays exactly 2. For a -phase gate the figure of merit is with as much transmission as possible; the phase condition is met exactly across the relevant range, and the whole cost is contrast.
That the conditional phase survives is consistent with the single-photon inversion above. The linear phase enters the two-photon state twice, so the flip at changes by , that is, not at all—and , which divides out , is invariant under a sign flip common to both photons. A -phase gate is a difference between one- and two-photon responses, so whatever is common to both cancels from it. That insensitivity is exactly what makes robust, and exactly why the penalty has nowhere to go except into the amplitude. Continuous-wave chiral nonlinear optics and all-optical control draw on the same interaction with coherent fields rather than Fock-state pulses [18, 19].
Dephasing distorts rather than scales.—It is tempting to summarize the above as an overall scale, writing the realized node map as with the target unitary and real. A normalized, phase-sensitive fidelity would then divide out and report identically, and the dephasing penalty would be pure loss: a rate multiplying the heralding probability, compounding as over cascaded nodes.
Importantly, no single exists, and Figure 3(c,Figure 3d) shows why. The one- and two-photon channels are suppressed at different rates. Normalized to their values, the single-photon transmission falls as , reaching zero at , while the resonant two-photon transmission falls only to of its value there (i.e., ). Were the two photons scattering independently the latter would track ; instead it exceeds that expectation by factors of , and at , , , and diverges from it at , where the node transmits two-photon amplitude while transmitting none at one photon. A single scalar cannot describe this.
Dephasing does two distinct things: a part common to all sectors, which is a pure scale, and a part that reweights the coherent photon-number sectors against one another, which is a genuine distortion of the target unitary. Only the first is invisible to a normalized fidelity. Neither is trainable: a network tunes phase shifters and splitting ratios, i.e., unitary parameters, whereas the missing weight has left the coherent subspace irreversibly into incoherent emission, and the reweighting is fixed by at the emitter, a rate over which the circuit has no handle. For architectures that cascade emitter nonlinearities [13, 14, 15], this is precisely the imperfection the central mechanism cannot address, and it should be accounted for as two independent penalties, not a single efficiency factor. A network-level study will be presented in another work.
The reweighting is directly measurable, and with apparatus that exists. Both and are coherent amplitudes referenced against an unscattered arm, i.e., precisely the quantity the Mach–Zehnder measurement of Ref. [12] measures. Sending first a one-photon and then a two-photon pulse through the same interferometer, on resonance and at fixed , and comparing the two-photon fringe contrast against the square of the one-photon contrast tests Figure 3(d) directly: a linear element, or an emitter acting on the two photons independently, gives unity, whereas the bound-state term gives the growing excess predicted here. No absolute calibration is needed, since both are normalized to their own values.
The standard factorization fails.—The two-photon intensity correlator, , is a different object than the phase observable above, and here the situation is worse. It is a second moment: both operators carry the dephasing noise, so there is nothing noise-free to factorize against. The standard construction inserts a vacuum projector between the emitter operators, which replaces an average of a product by a product of averages. That step is exact when the zero-excitation subspace is one dimensional; this is true for arbitrary optical loss, which is why the published scattering theory is unaffected, but a Markovian requires bath weight at zero frequency, and the dephasing sector is never vacuum.
The factorization fails outright, since it breaks a conservation law. Pure dephasing destroys no excitations, so exactly at every ; the factorized expression instead falls short by 16–81% over –1, the window in which the theory is wanted Figure 4(c) End Matter. The discarded weight is a covariance, , hence one-signed: the factorization always undercounts, and the error is first order in , so there is no safe weak-dephasing regime to retreat to. The missing pairs are not missing from the experiment—they are the incoherently re-emitted photons of Figure 1(b,Figure 1c), which a photon counter records and an interferometer does not. Per noise realization the output state is pure, so the exact replacement is a single line, , with the per-realization amplitude; the End Matter gives it in closed form and validates it pointwise.
Figure 1. (a) A two-level emitter chirally coupled to a waveguide, subject to pure dephasing at rate . A Fock pulse is split on a 50:50 beamsplitter into a noise-free reference arm and a signal arm containing the emitter; the arms are recombined and coincidences at times and give the interferometric correlator that measures the phase. (b),(c) One- and two-photon Fock pulses scatter into a coherent output carrying the shift (drawn inverted) whose amplitude shrinks with , plus an incoherent remainder that carries the missing weight. Photon number is conserved at every ; coherent amplitude is not.
Figure 4. Panels (a)–(c) are , panel (d) is . (a) Equal-time two-photon amplitude : lines the closed form, Eq. (3), circles the Fock ME. (b) The conditional phase sits at wherever the amplitude is appreciable, at every ; the curves are offset only by the crossing time, , , , , , in order of increasing . The step on the leading edge is a sign change of a real quantity passing through zero, not a rotation. (c) Integrated pair number: the exact value , reproduced as 2.013 at every , against the vacuum cut , which falls short by up to 81%. (d) The single-photon counterpart: the fraction of the transmitted photon flux the cut fails to account for. It is non-monotonic because strong dephasing decouples the emitter: , so the transmitted field approaches the unscattered input and the factorization becomes exact again. The deficit peaks near and decays; the pair deficit of (c) behaves the same way, with its maximum near , beyond the range plotted.
Conclusions.—We have derived the two-photon scattered amplitude of a chirally coupled two-level emitter with pure dephasing in closed form, Eq. (3), validated against an independent Fock-state master equation and against frequency-domain scattering theory at . The conditional phase is pinned at for every pure-dephasing rate examined, so pure dephasing destroys the two-photon amplitude rather than rotating the nonlinear phase. The one- and two-photon channels are attenuated unequally, so the dephased node does not simply pass less light: it changes the balance between the one- and two-photon channels, an error that a normalized fidelity does not register and that no phase setting can repair.
Two directions follow. First, chaining Eq. (3) over several nodes would give network-scale predictions without re-simulating the many-body dynamics, provided the cascade closes analytically even though each node sees an input already correlated by the previous one. Second, the white-noise model of pure dephasing, shared with all the cited work, is justified here. Phonon dephasing in quantum dots is non-Markovian [40, 41, 42, 43, 44, 45], but its correlation time, a few picoseconds, is far below the pulse durations of interest. At the acoustic sideband then reduces to a temperature-dependent Franck–Condon factor at two photons, a calculable prefactor that spectral filtering largely removes [45, 46]. Non-Markovian corrections matter mainly under strong driving, where pulse shaping can suppress them [33, 44, 47].
Acknowledgments
We acknowledge support from the Natural Sciences and Engineering Research Council of Canada (NSERC), the Canadian Foundation for Innovation (CFI), and Queen’s University, Canada.
References
End Matter
Fock-state master equation (Fock ME) benchmark.— No scattering treatment includes dephasing, so there is nothing in the literature to check Eq. (3) against. We therefore validate it against a Fock-state master equation solved under identical assumptions. For an -photon Fock pulse with coupling operator , define generalized operators , , obeying
with and ; the physical state is [48, 49, 50, 51]. Dephasing enters as one more Lindblad term: no factorization, no vacuum insertion, no restriction on . The labels count the photons dressing ket and bra, so the amplitudes entering the phase live in off-diagonal blocks; has a two-photon ket and a vacuum bra, hence , with and propagation by Eq. (5).
At the Fock ME reproduces the benchmarks of Ref. [52], including the peak and for , the node at , emitter population peaks and , the flux identity , and the bimodal-pulse results of Ref. [53]. Sum rules and hold to at every . Table I collects the comparisons; each entry is the largest relative deviation between the closed form and the reference calculation.
Derivation of the two-photon amplitude.— Projecting the Heisenberg–Langevin equation for onto Fock sectors closes a hierarchy labelled by the photon numbers of bra and ket. For one photon,
with and the initial population; for an emitter starting in this integrates to the quoted below Eq. (3). All the dephasing dependence sits in : each carries with , only the difference over the interval separating excitation from de-excitation appears, and the Gaussian average gives .
| route (regime) | agreement |
| Fock ME (time domain, all ) | – |
| analytic integral (, grid free) | 5 sig. figures |
| scattering theory () | worst |
| covariance core (Monte Carlo) | (MC floor) |
| 2D map (cells of peak) () | () |
Table 1. Validation of the closed-form two-photon amplitude.
Each in Eq. (2) is a single amplitude with the vacuum already in the bra, so no factorization is performed and the average is exact. The hierarchy closes from the top: is known outright, are driven by it, and by those. Substituting back and separating terms that factorize as a function of times a function of from the one term that does not yields Eq. (3): the factorizable part is , two independent scatterings, and the remainder is the bound-state term, whose exponential is the emitter coherence surviving between the two events.
Why the standard factorization fails.— Expanding gives terms, according to whether each operator is input field or emitter; the emitter-bilinear ones are conventionally factorized by inserting a vacuum projector between them—the “vacuum cut”, so called because between the two emitter operators the system must return to the zero-excitation subspace. That is exact whenever that subspace is one dimensional. Loss reservoirs are vacuum to excellent approximation, so the cut survives arbitrary ; we verify a population defect of at . A Markovian , however, needs bath weight at zero frequency, and the dephasing sector is never vacuum.
The correct object is a covariance. Take the all-emitter term, the only one needing the insertion on both sides. Within a realization the state is pure and the insertion exact, so it is , whereas inserting the projector after averaging gives . The difference is the covariance of across realizations, nonzero because both factors are driven by the same . The Gaussian average can be done exactly, giving a quadruple integral dressed by , with the total time over which a raising and a lowering leg co-propagate; recovers the cut. Two consequences follow that numerics alone would not give: the error is first order, , so there is no safe weak-dephasing regime; and the cut is a strict lower bound: off the diagonal, and on it the exact term is , which cannot fall below by Cauchy–Schwarz.
The sixteen terms are not in fact needed separately. Per realization the output state is pure, so the exact correlator and its factorized counterpart are
and Eq. (8) supersedes the term-by-term construction: the sixteen terms are just sorted by powers of . The simplest case is , where the difference between them reduces to the population defect , a variance, already published as in Ref. [34], App. C. We claim no novelty at ; the point is that the variance reading makes the sign manifest and generalizes.
Pulse duration.— The duration used throughout, , is close to the best available choice, and two quantities set it. Gate performance is known to depend strongly on pulse bandwidth [25, 54]; dephasing supplies a second, competing constraint. The bound-state share of peaks at –: a shorter pulse is broader than the linewidth, so most spectral weight sits where , while a longer one weakens the nonlinearity relative to the linear term. Dephasing robustness, by contrast, favors short pulses monotonically, since the damage accumulates with the time the emitter holds coherence: the coherent fraction at falls 0.63, 0.49, 0.43, 0.38, 0.27 at , 0.5, 1, 2, 4. At it is unity at every duration, as unitarity requires.
The noise model.—White-noise pure dephasing, , is an idealization shared here by the input/output theory, the Fock-state master equation, and the cited literature alike, and its adequacy rests on a separation of timescales. The phonon correlation time is a few ps, short against the pulse durations of interest, so at the acoustic sideband does not reshape the coherent response but multiplies it by a temperature-dependent Franck–Condon factor , becoming at two photons; this is a known prefactor rather than a structural change, most of which spectral filtering removes [45, 46]. What remains is small at picosecond durations and matters chiefly under intense driving, where pulse shaping suppresses it further [33, 44, 47]. Pure dephasing is thus the dominant channel here, which is why treating it exactly is the useful step.
Slow noise is a different matter. Spectral diffusion (the wandering of the emitter’s transition frequency as nearby charges rearrange) has a correlation time of microseconds to milliseconds, so each repetition sees an essentially static detuning drawn from a distribution rather than a fluctuating one. The measured signal is then an average over , not a larger , which matters for any interferometric measurement accumulated over many shots, where spectral wandering broadens the line without being pure dephasing at all. It is also straightforward to include, since Eq. (3) holds at arbitrary detuning: the observable follows by averaging over the detuning distribution, and the variance that white noise hides becomes directly measurable because different runs genuinely sample different detunings. Like pure dephasing, this costs amplitude rather than adding a phase error, and no phase setting recovers it.
References
- [1]Y. Shen and J.-T. Shen, Phys. Rev. A 92, 033803 (2015).
- [2]D. E. Chang, L. Jiang, A. V. Gorshkov, and H. J. Kimble, New Journal of Physics 14, 063003 (2012).
- [3]P. Lodahl, S. Mahmoodian, and S. Stobbe, Rev. Mod. Phys. 87, 347 (2015).
- [4]H. Le Jeannic, A. Tiranov, J. Carolan, T. Ramos, Y. Wang, M. H. Appel, S. Scholz, A. D. Wieck, A. Ludwig, N. Rotenberg, L. Midolo, J. J. García-Ripoll, A. S. Sørensen, and P. Lodahl, Nature Physics 18, 1191 (2022).
- [5]S. Mahmoodian, G. Calajò, D. E. Chang, K. Hammerer, and A. S. Sørensen, Phys. Rev. X 10, 031011 (2020).
- [6]P. Lodahl, S. Mahmoodian, S. Stobbe, A. Rauschenbeutel, P. Schneeweiss, J. Volz, H. Pichler, and P. Zoller, Nature 541, 473 (2017).
- [7]J. Petersen, J. Volz, and A. Rauschenbeutel, Science 346, 67 (2014).
- [8]I. Söllner, S. Mahmoodian, S. L. Hansen, L. Midolo, A. Javadi, G. Kiršanskė, T. Pregnolato, H. El-Ella, E. H. Lee, J. D. Song, S. Stobbe, and P. Lodahl, Nature Nanotechnology 10, 775 (2015).
- [9]R. J. Coles, D. M. Price, J. E. Dixon, B. Royall, E. Clarke, P. Kok, M. S. Skolnick, A. M. Fox, and M. N. Makhonin, Nature Communications 7, 10.1038/ncomms11183 (2016).
- [10]D. Martin-Cano, H. R. Haakh, and N. Rotenberg, ACS Photonics 6, 961 (2019).
- [11]M. J. Mehrabad, A. P. Foster, R. Dost, E. Clarke, P. K. Patil, A. M. Fox, M. S. Skolnick, and L. R. Wilson, Optica 7, 1690 (2020).
- [12]M. J. R. Staunstrup, A. Tiranov, Y. Wang, S. Scholz, A. D. Wieck, A. Ludwig, L. Midolo, N. Rotenberg, P. Lodahl, and H. Le Jeannic, Nature Communications 15, 7583 (2024), preprint arXiv:2305.06839.arxiv.org/abs/2305.06839
- [13]J. Ewaniuk, J. Carolan, B. J. Shastri, and N. Rotenberg, Advanced Quantum Technologies 6, 2200125 (2023), preprint arXiv:2208.06571.DOI
- [14]J. Ewaniuk, B. J. Shastri, and N. Rotenberg, Large-scale tree-type photonic cluster state generation with recurrent quantum photonic neural networks (2025), arXiv:2505.14628 [quant-ph].arxiv.org/abs/2505.14628
- [15]I. M. Boras Vazquez, J. Ewaniuk, and N. Rotenberg, Quantum photonic neural networks in time (2026), arXiv:2603.23798 [quant-ph].arxiv.org/abs/2603.23798
- [16]S. Mahmoodian, P. Lodahl, and A. S. Sørensen, Phys. Rev. Lett. 117, 240501 (2016).
- [17]D. Istrati, Y. Pilnyak, J. C. Loredo, C. Antón, N. Somaschi, P. Hilaire, H. Ollivier, M. Esmann, L. Cohen, L. Vidro, C. Millet, A. Lemaitre, I. Sagnes, A. Harouri, L. Lanco, P. Senellart, and H. S. Eisenberg, Nature Communications 11, 10.1038/s41467-020-19341-4 (2020).
- [18]C. Dufresne, A. Makowski, and N. Rotenberg, Chiral non-linear optics and optical control (2026), arXiv:2608.14871 [quant-ph].arxiv.org/abs/2608.14871
- [19]C. Dufresne, A. Makowski, and N. Rotenberg, Multicolor nonlinear chiral quantum optics: beyond phase (2026), arXiv:2608.14807 [quant-ph].arxiv.org/abs/2608.14807
- [20]A. McCaw, J. Ewaniuk, B. J. Shastri, and N. Rotenberg, Nanophotonics 13, 2951 (2024), arXiv:2312.10521.DOI
- [21]T. Lund-Hansen, S. Stobbe, B. Julsgaard, H. Thyrestrup, T. Sønder, M. Kamp, A. Forchel, and P. Lodahl, Phys. Rev. Lett. 101, 113903 (2008).
- [22]T. Pregnolato, X.-L. Chu, T. Schröder, R. Schott, A. D. Wieck, A. Ludwig, P. Lodahl, and N. Rotenberg, APL Photonics 5, 086101 (2020).
- [23]A. V. Kuhlmann, J. Houel, A. Ludwig, L. Greuter, D. Reuter, A. D. Wieck, M. Poggio, and R. J. Warburton, Nature Physics 9, 570 (2013).
- [24]F. T. Pedersen, Y. Wang, C. T. Olesen, S. Scholz, A. D. Wieck, A. Ludwig, M. C. Löbl, R. J. Warburton, L. Midolo, R. UppU, and P. Lodahl, ACS Photonics 7, 2343 (2020), https://pubs.acs.org/doi/10.1021/acsphotonics.0c00758.pdf
- [25]A. Nysteen, D. P. S. McCutcheon, M. Heuck, J. Mørk, and D. R. Englund, Phys. Rev. A 95, 062304 (2017).
- [26]H. Le Jeannic, T. Ramos, S. F. Simonsen, T. Pregnolato, Z. Liu, R. Schott, A. D. Wieck, A. Ludwig, N. Rotenberg, J. J. García-Ripoll, and P. Lodahl, Phys. Rev. Lett. 126, 023603 (2021).
- [27]P. Türtschmann, H. Le Jeannic, S. F. Simonsen, H. R. Haakh, S. Götzinger, V. Sandoghdar, P. Lodahl, and N. Rotenberg, Nanophotonics 8, 1641 (2019).
- [28]H. Pichler and P. Zoller, Phys. Rev. Lett. 116, 093601 (2016).
- [29]S. Arranz Regidor, G. Crowder, H. Carmichael, and S. Hughes, Phys. Rev. Res. 3, 023030 (2021).
- [30]R. Yanagimoto, E. Ng, L. G. Wright, T. Onodera, and H. Mabuchi, Optica 8, 1306 (2021).
- [31]O. Kaestle, R. Finsterhoelzl, A. Knorr, and A. Carmele, Phys. Rev. Res. 3, 023168 (2021).
- [32]A. Strathearn, P. Kirton, D. Kilda, J. Keeling, and B. W. Lovett, Nature Communications 9, 3322 (2018).
- [33]M. Richter and S. Hughes, Phys. Rev. Lett. 128, 167403 (2022).
- [34]M. Bundgaard-Nielsen, M. Kozma, S. Arranz Regidor, and S. Hughes, Phys. Rev. Research 8, 033109 (2026).
- [35]S. Arranz Regidor, M. Kozma, and S. Hughes, QwaveMPS: An efficient open-source Python package for simulating non-Markovian waveguide-QED using matrix product states (2026), code: https://github.com/SofiaArranzRegidor/QwaveMPS, arXiv:2602.15826 [quant-ph].arxiv.org/abs/2602.15826
- [36]Because γ0 and γφ enter the coherent amplitudes only through Γ, the coherent output of a dephased emitter is identical to that of an emitter with off-chip decay γ0 + γφ and no pure dephasing, at any photon number, since the jump terms of both Lindblad channels vanish in the coherent block. The two differ only in the incoherent remainder, which loss removes from the waveguide and dephasing returns to it.
- [37]Equation (3) is written for both photons in one mode. For the input A†[f1]A†[f2]|0⟩ with distinct envelopes f1, f2, the same derivation gives ψ′(t1,t2) = ψ′(t1)ψ′(t2) + 2γe−|ζt2−t1|C(1)1(tmin)C(1)2(tmin), with ψ′j = fj − i√γC(1)j and C(1)j driven by fj; Eq. (3) follows on setting f1 = f2 = f and dividing by √2, the normalization of |2f⟩. Since the amplitude is linear in the input, entangled two-photon inputs follow by superposition. Nothing in the dephasing average depends on the input, so the γφ dependence is unchanged.
- [38]ψ′ is real and C(1) purely imaginary, so Rcond is real and its argument can only be 0 or π.
- [39]While quantization of ϕcond is guaranteed by resonance and real f, whether Rcond is positive or negative is dependent on the strength of the nonlinear interaction.
- [40]A. J. Ramsay, A. V. Gopal, E. M. Gauger, A. Nazir, B. W. Lovett, A. M. Fox, and M. S. Skolnick, Phys. Rev. Lett. 104, 017402 (2010).
- [41]C. Roy and S. Hughes, Phys. Rev. Lett. 106, 247403 (2011).
- [42]J. Iles-Smith, D. P. S. McCutcheon, A. Nazir, and J. Mørk, Nature Photonics 11, 521 (2017).
- [43]T. Grange, N. Somaschi, C. Antón, L. De Santis, G. Coppola, V. Giesz, A. Lemaître, I. Sagnes, A. Auffèves, and P. Senellart, Phys. Rev. Lett. 118, 253602 (2017).
- [44]C. Gustin and S. Hughes, Phys. Rev. B 98, 045309 (2018).
- [45]E. V. Denning, J. Iles-Smith, N. Gregersen, and J. Mørk, Opt. Mater. Express 10, 222 (2020).
- [46]L. Dusanowski, C. Gustin, S. Hughes, C. Schneider, and S. Höfling, Nano Letters 22, 3562 (2022).
- [47]G. Crowder, L. Ramunno, and S. Hughes, Phys. Rev. A 113, 033709 (2026).
- [48]K. M. Gheri, C. Ellinger, T. Pellizzari, and P. Zoller, Fortschr. Phys. 46, 401 (1998).
- [49]B. Q. Baragiola and J. Combes, Phys. Rev. A 96, 023819 (2017).
- [50]J. Combes, J. Kerckhoff, and M. Sarovar, Advances in Physics: X 2, 784 (2017).
- [51]A. H. Kiilerich and K. Mølmer, Phys. Rev. Lett. 123, 123604 (2019).
- [52]S. Arranz Regidor, M. Kozma, and S. Hughes, Phys. Rev. Res. 8, 033260 (2026).
- [53]S. Arranz Regidor, A. Knorr, and S. Hughes, Phys. Rev. Res. 7, 023295 (2025).
- [54]A. Nysteen, P. T. Kristensen, D. P. S. McCutcheon, P. Kaer, and J. Mørk, New Journal of Physics 17, 023030 (2015).



