The SLH framework for modeling quantum input-output networks

I. INTRODUCTION

• The review motivates SLH as a physics-oriented account of quantum input-output networks whose modularity and coherent interconnects support control-theoretic analysis of quantum and semiclassical systems.It emphasizes physical intuition over fully rigorous mathematical proofs to make the methodology accessible to a wider community.Source: §I · pp. 2, 4
• SLH represents localized quantum components by (S, L, H) triples and composes them into reduced network models, enabling systematic equations of motion and partially automated design and analysis.The framework supports arbitrary interconnections, including feedback, while preserving a common representation for components and networks.Source: §I · pp. 2–3

A. History

• Input-output theory and cascaded quantum systems provide SLH’s foundations, combining Gardiner-Collett field relations, cascaded connections, and Hudson-Parthasarathy quantum stochastic differential equations.Gough and James later developed SLH to compose and analyze arbitrary, not necessarily linear, components.Source: A. History · pp. 4–5

B. Applications and advantages

• Coherent feedback can outperform measurement-based feedback because it processes both non-commuting output quadratures without the additional noise caused by measuring only one quadrature.This advantage supports tasks including backaction-evading measurements, quantum nondemolition variables, and decoherence-free subsystems.Source: B. Applications and advantages · pp. 5–6, A. History · p. 5
• QIONs have been experimentally demonstrated in optical and superconducting microwave systems, but itinerant-field networks remain more loss-sensitive and difficult to implement than direct coherent interactions.Their prospective value combines modularity and flexibility with the speed, integrability, and fundamental advantages of coherent control.Source: B. Applications and advantages · p. 6

A. Input-output relations

• Quantum input-output relations begin with a localized system coupled to a bosonic field, then apply weak coupling, the rotating-wave approximation, and the Markov approximation to obtain a memoryless broadband interaction.The Markov approximation is typically reliable for weak, narrowband system-bath coupling and dynamics slow compared with 1/ζ.Source: A. Input-output relations · pp. 7–8
• The derivation assumes weak, linear system-field coupling and restricts coupling operators to off-diagonal transitions in the system-energy basis.Linear coupling is justified when the interaction is much weaker than system and bath energies; diagonal or degenerate components require more general Markov treatments because they invalidate extending frequency integrals to negative values.Source: A. Input-output relations · pp. 8–9
• The quantum Langevin equation describes how an input field bin(t), modeled as singular quantum white noise, drives the dynamics of an arbitrary system operator.The input field is defined from initial bath modes and has delta-function commutation relations, motivating smoother noise increments introduced later.Source: A. Input-output relations · p. 9
• The input-output relation maps incident field modes to scattered output modes and determines output-field properties from the input and the system operator c(t).The same boundary condition can be represented in space-time, where field excitations to the left of the localized system are inputs and those to the right are outputs.Source: A. Input-output relations · p. 9
• For a single-mode resonator, the framework gives the cavity-mode equation of motion and input-output relation using the detuning, decay rate, and coupling operator L = √γc.The example applies the general relations to an empty optical or microwave resonator coupled to a guided wave or transmission line.Source: A. Input-output relations · p. 10
• The time-ordered unitary propagator generates the Heisenberg evolution and is validated by deriving a generator that reproduces the quantum Langevin equation.The proof uses an identity for the commutator of the input process with the propagator and recovers the Langevin equation exactly.Source: A. Input-output relations · pp. 10–11

Setup

• Cascading connects two cavities through a unidirectional field, with a circulator preventing back scattering from the second cavity to the first.The setup assumes weak coupling and Markov approximations, and distinguishes cascading from simple Hamiltonian coupling.Source: B. Cascaded systems · p. 11, Figure 3 · p. 12

Zero-delay effective model

• When propagation delay is negligible, the inter-cavity field degrees of freedom can be eliminated, yielding effective Hamiltonian Heff and coupling operator Leff for the composite system.The approximation requires τ to be much shorter than the relevant dynamical timescales, although finite delays can matter for centimeter-scale distances and gigahertz dynamics.Source: B. Cascaded systems · pp. 12–13

Directional dynamics

• Cascaded dynamics break time-reversal symmetry: the second system is affected by the first, while the first is unaffected by the second.For operators in the first cavity, terms involving the second cavity vanish; operators in the second can depend on both cavities and the input field.Source: B. Cascaded systems · p. 12

Framework motivation

• The cascaded-system derivation motivates SLH rules for more complex interconnected networks, including feedback connections that extend the range of networked quantum dynamics.The framework treats the two-cavity cascade as one effective component rather than retaining the intermediate field explicitly.Source: B. Cascaded systems · pp. 12–13

Example III.2

• In the empty-cavity example, the first mode drives the second mode’s evolution, while the output field contains information from both cavities.This follows by substituting the effective Hamiltonian and coupling operator into the input-output model.Source: B. Cascaded systems · p. 13

Quantum noise increments

• Singular broadband input fields are regularized through time-integrated quantum noise increments whose commutation relations and vacuum moments resemble, but are not identical to, classical Wiener increments.Second-order increment products can be proportional to dt under the vacuum state, motivating quantum stochastic integration.Source: §IV · pp. 13–14

Quantum stochastic integration

• Itō and Stratonovich integrals differ for quantum stochastic increments, so every QSDE must specify its interpretation.Unlike ordinary calculus, the two definitions are inequivalent because quantum increments remain singular in the continuum limit.Source: §IV · p. 14

Stratonovich and Itō forms

• Physically derived QSDEs correspond to Stratonovich form under the white-noise limit, while their Itō form includes an additional Itō correction.The modified product rule contains the correction term dX(t)dY(t), reflecting the difference from standard calculus.Source: §IV · p. 14
• The reviewed system-operator and propagator QSDEs are interpreted in Stratonovich form but converted to Itō form because Itō equations are easier to manipulate analytically and numerically.The SLH framework is developed in Itō form, so the remainder of the treatment works exclusively with Itō QSDEs.Source: §IV · p. 14

Example IV.1: Deriving equations of motion in Itō form using quantum stochastic calculus

• Quantum stochastic calculus derives Heisenberg equations by applying the vacuum Itō table and retaining second-order differentials, recovering the single-sided resonator’s earlier equation of motion in Itō form.Terms are eliminated according to the increment-product rules, then system and field operators are normally ordered before the final expression is obtained.Source: §IV · p. 15

V. GENERAL QUANTUM INPUT-OUTPUT NETWORKS AND THE SLH FRAMEWORK

• The SLH framework enables large networked quantum systems to be constructed through simple algebraic manipulations, extending the cascaded approach developed by Gough and James.Source: §V · p. 16
• Local quantum systems are described by QSDE propagators parameterized by (S, L, H), which can be composed according to network connectivity.From the combined propagator, one derives network Heisenberg equations, input-output relations, and master equations.Source: §V · p. 16, A. The SLH time evolution operator and… · p. 17

A. The SLH time evolution operator and SLH triple

• The SLH time-evolution operator applies under weak, linear coupling and Markov approximations, generalizing earlier unitary propagators for local components and cascaded systems.Source: A. The SLH time evolution operator and… · p. 16
• In the SLH triple, L couples the localized system to external fields, H is its Hamiltonian, and S describes direct field scattering, including phase shifts.When S = I, the equation reduces to the earlier single-scattering form; S ≠ I captures elements such as mirrors.Source: A. The SLH time evolution operator and… · p. 16
• For n input-output modes, L becomes a coupling vector and S an operator-valued unitary scattering matrix satisfying S†S = SS† = I.The modes may represent distinct polarization, spatial, frequency, optical, microwave, or phononic fields.Source: A. The SLH time evolution operator and… · pp. 16–17
• Multimode gauge increments dΛij represent direct photon scattering from mode j to mode i without internal energy exchange, while Sij encodes its system-dependent effect.Diagonal terms describe phase shifts, and the scattering operators obey the required unitarity constraints.Source: A. The SLH time evolution operator and… · p. 17
• Specifying (S, L, H), together with initial component and field states, completely determines all network properties because Eq. (52) fixes their coupling and time evolution.Source: A. The SLH time evolution operator and… · p. 17

Framework and assumptions

• The SLH composition rules algebraically combine component triples and are the heart of the framework.They apply when localized-component interactions are Markovian, connecting fields propagate in a dispersionless linear medium with negligible delay, and network inputs are vacuum; non-vacuum fields require later extensions.Source: B. SLH composition rules · pp. 18–19

Basic composition products

• The series product cascades matched output-input fields, while concatenation groups independent components in parallel and direct coupling adds an intersystem Hamiltonian.The series product is generally noncommutative because directional fields induce coupling-operator terms in the effective Hamiltonian; direct coupling places the interaction according to the stated convention.Source: B. SLH composition rules · pp. 19–20

Feedback reduction

• Feedback reduction eliminates internal output-to-input links and produces a reduced SLH triple whose result depends on which ports are connected.Its coupling terms can represent repeated, indeterminate scatterings through the feedback loop, and the series product is a special case of feedback reduction.Source: B. SLH composition rules · pp. 20–21

Network construction

• Concatenation followed by feedback reduction is sufficient to compose arbitrary networks from any number of components.The general procedure first treats all components as independent, then applies feedback reduction to implement every internal connection.Source: B. SLH composition rules · p. 20

Port matching and rewiring

• Padding elements allow cascading networks with unequal port counts, while permutation components reroute channels so nonmatching input-output mappings can be represented.For multiple feedback eliminations, elimination order is immaterial, but ports must be block-contiguous and indices carefully tracked; naive application can fail without the required permutations.Source: B. SLH composition rules · pp. 20–22

Examples

• Examples apply the rules to cascaded, parallel, directly coupled, and feedback-connected cavities, with feedback reducing a two-port resonator to one port and changing its effective coupling and detuning.A permutation with S operator P[3,1,2] maps the specified third, first, and second outputs of one component to the first, second, and third inputs of another.Source: B. SLH composition rules · pp. 23–24

• Markovian propagator factorization and commutation with past input fields enable infinitesimal expansion of output increments into quantum stochastic differential equations.Input processes use temporal-mode labels, and output increments are defined from integrated output fields.Source: C. Network Heisenberg equation of… · pp. 24–25
• The general multiple-input/output relations express output field increments as scattered input fields plus contributions from the network’s localized components.The corresponding gauge-process relation combines input photons with localized-component contributions during each time increment.Source: C. Network Heisenberg equation of… · p. 25
• For a single port, the input-output relation reduces to dB_out(t) = S(t)dB_in(t) + L(t)dt, recovering the standard relation for L = √γa and S = I.The derivation uses that the network propagator at time t commutes with increments at time t because it depends only on earlier increments.Source: C. Network Heisenberg equation of… · p. 25
• The framework also derives Heisenberg equations of motion for arbitrary localized-system operators, with compact and single-port forms expressed using quantum stochastic calculus.The operators are written in the Heisenberg picture, and an alternative quantum-flow notation makes this picture explicit.Source: C. Network Heisenberg equation of… · pp. 26–27
• In a two-cavity network interrupted by a beamsplitter, cascading drives the second cavity through the first cavity and both input fields, while output fields contain mode and input-field superpositions.Padding components maintain field indexing, and the beamsplitter models transmission-line loss in the example network.Source: C. Network Heisenberg equation of… · pp. 27–28

Master equation derivation

• Tracing over vacuum input-output fields yields a Markovian master equation for the localized density matrix ρ(t), averaging over propagating-field effects.The localized state is obtained by a partial trace over all input-output field Fock spaces, with initial field state assumed to be vacuum.Source: D. Master equation description · p. 28
• Factoring the initial state into localized and vacuum-field components and applying the Itō table produces the master equation for a general SLH network G = (S, L, H).The resulting equation is deterministic because the stochastic quantities have been averaged over.Source: D. Master equation description · pp. 28–29

Conditioned dynamics

• Monitoring selected output fields instead of tracing over all of them leads to conditioned stochastic master, quantum-trajectory, or quantum-filtering equations.The review cites extensive introductions and primary literature on these equations but does not discuss the topic further.Source: D. Master equation description · p. 29

Input-field assumptions

• Under vacuum input, the scattering matrix S does not appear in the reduced master equation; with non-vacuum inputs, the equation can depend on S.The non-vacuum case is deferred to Sec. VII.A and Example VII.1.Source: D. Master equation description · p. 29

Definition and scope

• The section restricts attention to experimentally accessible linear quantum systems formed from collections of harmonic degrees of freedom, especially in the optical regime studied in quantum optics.Each mode is characterized by bosonic annihilation and creation operators obeying canonical commutation relations.Source: §VI · p. 29
• Linear QIONs comprise harmonic modes with quadratic Hamiltonians and linear couplings to external propagating fields; Gaussian measurements preserve the linear description.Examples include homodyne detection of arbitrary quadratures and heterodyne detection.Source: §VI · p. 29

Representations and control

• It formally specifies linear QIONs, develops representations alternative to the SLH triple, and reviews classical-control techniques transferred to quantum linear systems.Section VI.C presents selected techniques and results from this control-theoretic framework.Source: §VI · p. 29

A. Passive linear quantum networks

• Passive networks comprise components such as beam-splitters, phase shifters, and empty cavities, represented by SLH triples with scalar S, linear L, and photon-number-conserving quadratic H.These restrictions define the passive linear subclass of quantum input-output networks.Source: A. Passive linear quantum networks · p. 29
• The passive-network internal dynamics satisfy forced linear differential equations, while outputs depend linearly on inputs and internal annihilation operators through matrices A, B, C, and D.For linear QIONs, specifying these matrices is equivalent to specifying the network through an SLH triple.Source: A. Passive linear quantum networks · pp. 29–30
• Unlike classical ABCD systems, quantum A, B, C, and D matrices are interdependent because unitary evolution and the uncertainty principle constrain their allowable values.The SLH-to-ABCD relations can also be reversed to recover an SLH representation from a linear system.Source: A. Passive linear quantum networks · p. 30
• The passive transfer-function matrix Ξ(s) is sufficient to specify the network’s input-output behavior in the Laplace domain.It is obtained by solving the linear differential equation and relating transformed input and output fields.Source: A. Passive linear quantum networks · p. 30

B. Active linear quantum networks

• Active networks include energy-nonconserving elements such as squeezers and amplifiers, requiring a doubled state vector containing annihilation and creation operators.Their SLH triples retain scalar S but allow L to depend linearly on both operator types and H to be a general quadratic Hamiltonian generating symplectic transformations.Source: B. Active linear quantum networks · p. 30
• Active networks also have linear equations and input-output relations, but their doubled-up ABCD matrices remain subject to additional quantum constraints and define a doubled-up transfer function.Quadrature variables provide an alternative representation, while linear quantum networks preserve Gaussian states.Source: B. Active linear quantum networks · p. 31
• In the coherent-feedback example, connecting an optical parametric oscillator to a beam-splitter rescales the cavity decay κ by l2 ≤1 and enhances deamplification as η →0.The opposing quadrature is amplified by the same amount, so the incident field becomes increasingly squeezed while preserving phase-space area.Source: B. Active linear quantum networks · pp. 31–33, Figure 8 · p. 32

C. Survey of results regarding linear quantum networks

• Linear quantum networks inherit Hurwitz stability from classical systems, while controllability and observability are characterized by analogous matrix-rank conditions [89, Chapter 6][108].Source: C. Survey of results regarding linear… · p. 33
• A linear quantum system is physically realizable only under specified conditions, which ensure preservation of the canonical commutation relations.Source: C. Survey of results regarding linear… · p. 33
• Quantum feedback research extends classical control through realizable LQG and H∞ designs, with rank-constrained LMI formulations and robustness to model uncertainties [30].The H∞ formulation incorporates controller realization conditions, guaranteeing that the coherent feedback controller can be implemented physically.Source: C. Survey of results regarding linear… · pp. 33–34
• Arbitrary linear quantum systems can be synthesized from cascaded harmonic oscillators with direct interactions, while later schemes remove direct couplings or realize transfer functions using purely cascaded networks [110][111][112].Additional results address alternative passive-network topologies, balanced-truncation model reduction, and system identification [113; 114][115].Source: C. Survey of results regarding linear… · p. 34

VII. EXTENSIONS TO THE SLH FRAMEWORK

• SLH extensions model experimental phenomena and imperfections by composing standard building blocks into customized components while preserving modular network structure.The review emphasizes that these extensions are active and non-exhaustive, often approximating complex behavior through freely propagating fields interacting with component sequences.Source: §VII · p. 34

A. Non-vacuum input states via source models

• Because standard SLH composition assumes vacuum inputs, non-vacuum fields require extensions such as source models that replace the field with an explicitly modeled source component.This source component is typically a minimal idealized apparatus producing the desired field state, and the approach originated in cascaded-system modeling of arbitrary light statistics [14].Source: A. Non-vacuum input states via source… · p. 34
• A source model chooses an SLH triple GS = (SS, LS, HS) and initial state ρS(0), drives it with vacuum, and cascades it with G1 so G1 behaves as if driven by field state ρφ.Figure 9 depicts this fictitious engineered source system and its initial state at t = 0.Source: A. Non-vacuum input states via source… · p. 34, Figure 9 · p. 35

1. Coherent states

• Coherent states describe pulsed laser light through a wave-packet profile ξ(t), with mean photon number n̄ = |α|^2∫ds|ξ(s)|^2 and vacuum or single-mode states as special cases.The displacement operator provides the corresponding field construction, and α(t) ≡ αξ(t).Source: §1 · p. 35
• A simple coherent-displacement source reproduces the desired input-output behavior more efficiently than a cavity model, while the physical cavity representation generalizes to other field states.The cavity source uses vacuum input and an initially prepared cavity state; both models produce the same output behavior.Source: §1 · pp. 35–36
• Cascading a coherent source into an arbitrary component yields the coherent-state master equation, where the S operator appears explicitly under non-vacuum driving.The equivalent Lindblad-form derivation includes coherent drive terms and an |α(t)|^2 scattering contribution.Source: §1 · pp. 35–37

2. Finite-bandwidth squeezed states

• Finite-bandwidth squeezed light can be modeled by a degenerate OPO cavity, whose output is quadrature squeezed over bandwidth γ and whose Y quadrature is squeezed.The same construction extends to two-mode squeezed states and, by tracing out one mode, thermal states.Source: §2 · p. 37

3. Fock and N-photon states

• Continuous-mode Fock states distribute photons across a temporal wave packet, producing temporal correlations that make the driven system non-Markovian but representable through a larger Markovian source model.Single-photon cascaded models were generalized to superpositions or mixtures of single photon and vacuum, and to arbitrary Fock states.Source: §3 · pp. 37–38
• General N-photon states allow arbitrary joint spectral or temporal distributions, but their source models are fairly complicated and include continuous matrix product states.Continuous-mode N-photon states belong to this broader family, for which corresponding master equations have been derived.Source: §3 · p. 38
• N photons in distinct, possibly overlapping wave packets can be mimicked by a multimode cavity with N time-dependent couplings to one input-output field.This is the tractable special case associated with photon-gun-like inputs.Source: §3 · p. 38

4. Cat states

• Cat states are normalized superpositions of distinct continuous-mode coherent states, with source constructions based on either an n-level qudit or a multimode cavity.The qudit construction uses level projectors with time-dependent couplings, while the initial qudit state is chosen from the superposition weights and coherent-state overlaps.Source: §4 · p. 38

B. Alternatives to source models

• Non-vacuum inputs can be handled either by source models or by decomposing fields into tractable bases, where ordinary SLH composition rules apply.The reviewed bases include Fock, N-photon, selected multiphoton, and cat states; source models may nevertheless be difficult to construct for some QION inputs.Source: B. Alternatives to source models · p. 38

1. Simulation in a Fock basis

• Fock-basis simulation represents SLH-network dynamics through coupled master equations for generalized state matrices, initialized from the system state and zero off-diagonal blocks.The method also evaluates system observables and output quantities such as mean photon flux, and extends to multiple modes and spectrally entangled inputs.Source: §1 · p. 39

2. General gaussian input states

• Gaussian states include coherent, squeezed, and thermal fields and are characterized by first and second moments, with time-varying mean α(t) and covariance parameters N and M.When M = 0, the field is thermal; nonzero M indicates squeezing, and N(N + 1) = |M|^2 means squeezing without thermal photons.Source: §2 · pp. 39–40
• For a single localized component, Gaussian-input dynamics modify the system’s operator equation and master equation through N-, M-, and α(t)-dependent terms, while input-output relations remain unchanged.The resulting master equation can be diagonalized into Lindblad form and admits homodyne and heterodyne unravellings; multiple orthogonal input-output modes generalize by adding mode indices.Source: §2 · pp. 40–41
• The Gaussian-input extension does not generally support arbitrary scattering matrices: it can effectively model simple static beamsplitters, but not general S ≠ I scattering dynamics.This limitation reflects a fundamental obstacle to incorporating pure scattering dynamics with arbitrary non-vacuum Gaussian fields.Source: §2 · p. 41
• Ref. [70] extends standard SLH series and feedback rules to arbitrary Gaussian inputs, but the resulting network triples correspond to Stratonovich rather than input-state-independent It¯o dynamics.An It¯o representation makes the L operators depend on the Gaussian field state, conflicting with SLH modularity because component descriptions should encode intrinsic properties independently of their inputs.Source: §2 · p. 41

C. Emission and propagation losses

• Losses are represented by fictitious modes or ports with vacuum inputs, followed by tracing over the non-guided modes.Cavity loss uses a concatenated fictitious port, while distributed waveguide or free-space loss is captured with one or more fictitious beam splitters.Source: C. Emission and propagation losses · pp. 41–42
• Fictitious loss channels must match the physical emission pathways, because an atom may emit directly into non-guided modes, through a cavity, or through both mechanisms.The modeling choice therefore depends on whether loss is direct, cavity-mediated, or jointly mediated.Source: C. Emission and propagation losses · p. 42

D. Circulators

• A symmetric lossless three-port circulator is modeled by a unitary scattering matrix, while loss is added with fictitious beam splitters and imperfections through reflection and wrong-port transmission.For a symmetric imperfect circulator, the coefficients obey |t|^2 + |r|^2 + |b|^2 = 1 and rt* + tb* + br* = 0; desirable operation has |t| ≫ |r|, |b|.Source: D. Circulators · pp. 42–43
• Finite-bandwidth circulators are modeled with coupled-cavity SLH systems that approach the ideal scattering model near resonance, with bandwidth increasing as |γ| increases.For ϕ = −π/2 and t = γ/2, adiabatic elimination reduces the model to the ideal three-port form up to absorbable phases.Source: D. Circulators · p. 43

E. Bi-directional waveguides, back-reflections, and finite length propagation

• Counter-propagating waveguides are modeled with separate right- and left-moving input-output modes, phase shifts, and SLH composition rules that preserve propagation direction.The construction produces direction-dependent coupling phases and supports arbitrary numbers of cascaded impurities; more complicated topologies require feedback reduction.Source: E. Bi-directional waveguides,… · pp. 43–45
• Finite propagation distance L is approximated by inserting a phase shift exp[iφ], with φ = Δ_cL/ν, only when γL ≪ ν or the delay is negligible relative to system dynamics.For delays outside this regime, the phase-shift treatment is insufficient and time-delayed propagation must be modeled separately.Source: E. Bi-directional waveguides,… · p. 43, Figure 10 · p. 44
• Counter-propagation generates a Hamiltonian coupling proportional to sin(φ), while co-propagation instead adds an imaginary anti-Hermitian term proportional to cos(φ).Different phase shifts can be assigned to the two propagation directions, and the co-propagating case can arise in chiral quantum optics or through circulators [147].Source: E. Bi-directional waveguides,… · pp. 44–45

Framework

• Adiabatic elimination replaces fast SLH dynamics with a slow node on a projected subspace when the scaling parameter k approaches infinity.Operators scaling as k^2 generate fast dynamics, operators independent of k generate slow dynamics, and operators scaling as k couple the two timescales.Source: F. Model reduction by adiabatic… · pp. 45–46

Conditions

• The projection P0 must identify the slow subspace H0, while P1 = I − P0 identifies fast dynamics and must satisfy structural invertibility conditions.Choosing P0 is aided by physical insight and is often associated with the ground-state subspace, with P1 representing excited states.Source: F. Model reduction by adiabatic… · p. 46

Applications and limitations

• Eliminating fast components can reduce QION model complexity and simulation costs, but scaling coherent-state input amplitudes can introduce mathematical complications requiring modified procedures; [4, Sec. II C].Earlier optical-feedback applications used less rigorous elimination, while later work developed the SLH formulation and extensions.Source: F. Model reduction by adiabatic… · pp. 46–47

Network composition

• Adiabatic elimination commutes with SLH network composition: eliminating nodes before concatenation, series, and feedback products gives the same SLH parameters as eliminating after composition.This equivalence preserves the network dynamics along both composition pathways shown in Fig. 11.Source: Figure 11 · p. 46, F. Model reduction by adiabatic… · p. 47

Cavity-QED example

• In the cavity-QED example, the limit κ, g ≫ γ with projection onto the joint ground state yields an effective model with no internal Hamiltonian or effective couplings.Fast cavity and atomic excitations decay immediately, restricting internal dynamics to |0r, 0a⟩.Source: F. Model reduction by adiabatic… · pp. 47–48
• The effective cavity-QED input-output relations reflect the guided field with an additional π phase shift, while the unmonitored radiation mode acquires no phase shift.This corresponds to an on-resonant response for the guided probe and no appreciable dynamical effect on radiation modes as γ vanishes.Source: F. Model reduction by adiabatic… · p. 48

Continuum-limit approach

• Distributed transformations are modeled as many discrete components implementing infinitesimal transformations, followed by a continuum limit of the cascaded SLH model.This extends the framework beyond networks of discrete components to transformations distributed in space.Source: G. Modeling distributed transformations · pp. 48–49
• A continuum SLH model emerges by cascading thin slices or cavities across a medium and replacing spatially indexed sums with integrals as N →∞.Figure 12 illustrates the transition from cascaded cavities with varying decay rates and resonance frequencies to a continuous-medium model.Source: Figure 12 · p. 49, G. Modeling distributed transformations · p. 49

Gradient echo memories

• For gradient echo memories, thin atomic-ensemble slices are treated as cascaded cavities whose weak-excitation interactions are approximated by coherent exchanges with bosonic modes.The slices contain atoms with different detunings, and their thinness excludes emission followed by re-absorption within a single slice.Source: G. Modeling distributed transformations · p. 49

Applications

• Laplace-transform solutions of the resulting linear continuum model have been used to study gradient echo memories, single-photon production, and cross-phase modulation in two gradient echo memories.The output field is characterized through the network input-output relation after solving for the local continuum modes.Source: G. Modeling distributed transformations · pp. 49–50

Outlook

• Modeling material properties through continuum limits of SLH networks remains relatively unexplored but has significant potential.Source: G. Modeling distributed transformations · p. 50

H. SLH and scattering theory

• SLH connects asymptotic input and output field states through the unitary scattering matrix, while input-output operators provide its frequency-domain representation.The relation uses Møller operators and assumes the network components are in their ground state in the asymptotic past and future.Source: H. SLH and scattering theory · pp. 50–51
• A single-photon wavepacket emerges with a profile deformed by the scattering interaction, obtained by applying the S-matrix to the input wavepacket.The same SLH-based approach extends to two-photon, multimode, and coherent-state scattering calculations.Source: H. SLH and scattering theory · pp. 51–52
• SLH can calculate scattering matrices for arbitrary networks of quantum components, including finite-distance propagation phases and delays, although bound states can complicate the scattering problem.The framework has been applied to complex scattering calculations, while finite spatial distances and propagation delays are incorporated through propagation-dependent phases.Source: H. SLH and scattering theory · p. 52

I. Dispersive propagation

• Standard SLH assumes negligible dispersion between localized components, an assumption that can fail in integrated waveguides because of waveguide and material dispersion.These effects can be reduced through waveguide engineering, but eliminating dispersion entirely can be difficult.Source: §I · p. 52
• Dispersive propagation acts as a delay followed by convolution with a channel transfer function, producing frequency-dependent interaction modulation and invalidating the Markov approximation.For quadratic dispersion, this transfer function is modeled using a complex Gaussian.Source: §I · p. 53
• Dispersive propagation is fundamentally incompatible with the standard SLH Markov model, although distributed transformations can in principle be represented by many SLH components or larger Markovian embeddings.The incompatibility arises because SLH assumes free propagation except for localized interactions, whereas dispersion transforms propagating fields throughout the channel.Source: §I · p. 53
• Fictitious localized components can approximate quadratic dispersion when γ_d, Δ_d ≪ω_c and τ_p ≪τ_d, but this approximation fails for feedback loops.Higher-order dispersion can be modeled with additional fictitious components, yet no extension currently handles arbitrary dispersive propagation fully within SLH.Source: §I · p. 53

J. Time delayed field propagation

• Zero-delay propagation is central to SLH, but physically large networks and coherent feedback can require explicit treatment of finite delays; three approaches offer partial solutions.These approaches use fictitious SLH components, cascades driven by past outputs, or explicit MPS representations of in-loop fields.Source: J. Time delayed field propagation · p. 54, Figure 14 · p. 55

1. Approach 1: introduce fictitious SLH components

• Fictitious SLH networks approximate linear passive delay sub-networks by realizing fitted transfer functions with cavities, with accuracy improving as more poles and zeros are included.The resulting model remains compatible with localized SLH components and zero-delay interconnections, but increases network size.Source: §1 · pp. 54–55

2. Approach 2: cascades from the past

• Grimsmo’s super-operator product-state method represents delayed feedback as cascades of identical systems driven by past outputs, though it requires auxiliary components and is not yet integrated with SLH.Its numerical implementation is being developed in QuTiP, and transfer-tensor methods can approximate long-time dynamics.Source: §2 · p. 55

3. Approach 3: explicit representation of the in-loop fields

• Pichler and Zoller’s MPS method explicitly evolves system, input, output, and in-loop fields, allowing arbitrary delays but becoming infeasible as network size increases.It trades SLH’s elimination of intermediary fields for direct access to subsystem observables and correlations.Source: §3 · pp. 55–56

K. Software for automated modeling

• QHDL and QNET automate hierarchical SLH modeling from textual or graphical network descriptions, including triple derivation, simulation, symbolic analysis, component libraries, and circuit visualization.They have supported analyses of quantum memories, large-scale nanophotonic logic, and a coherent nonlinear superconducting microwave experiment,,.Source: K. Software for automated modeling · p. 56

VIII. INTEGRATED IMPLEMENTATIONS OF QIONS

• Integrated QIONs face higher loss, distributed material and waveguide effects, and difficult circulator fabrication, making direct SLH application non-trivial in silicon photonics and superconducting circuits.Some issues are addressed by SLH extensions, but platform-specific limitations remain.Source: §VIII · p. 56

A. Integrated quantum coherent networks in silicon photonics

• Silicon photonics requires models for dispersion, scattering, and nonlinear absorption: engineered low-dispersion waveguides can preserve SLH validity, whereas nonlinear Raman and Brillouin scattering generally exceed standard Markovian SLH modeling.Two-photon absorption and free-carrier effects can be incorporated through distributed nonlinear models, but the resulting state spaces are computationally large and may require semiclassical approximations.Source: A. Integrated quantum coherent… · p. 57

B. SLH and superconducting microwave systems

• SLH effectively models many superconducting microwave networks with resonators, couplers, and transmission lines, but Markov and free-field assumptions break down for lumped-element and near-field circuits.Bridging accurate lower-level circuit models with simpler SLH descriptions remains an open challenge, alongside non-Markovian modeling, scalable simulation, and nonlinear-network synthesis.Source: B. SLH and superconducting microwave… · p. 58, §IX · pp. 58–59

Basic optical and cavity components

• The appendix represents phase shifters, beam splitters, coherent drives, and cavities—including Kerr, Fabry–Perot, and cross-Kerr variants—within SLH notation.Beam-splitter scattering matrices must satisfy S†S = I; cavity models specify decay rates, detunings, and nonlinear Hamiltonian terms.Source: Appendix A · p. 60

Parametric components

• Degenerate OPOs and two-mode squeezing are modeled using cavity SLH triples with bandwidth, nonlinear-drive, pump, and rotating-frame parameters.The two-mode squeezing model uses a classical undepleted pump and a transformation to a frame rotating at half the pump frequency.Source: Appendix A · p. 60

Optomechanical systems

• Optomechanical SLH models include radiation-pressure coupling and a linearized version, with cavity and mechanical detunings, thermal-bath coupling, and phonon occupation.Approximations, detuning choices, and frame changes can produce cooling, heating, and QND-measurement interactions; extensions are cited in Refs..Source: Appendix A · p. 61

Atom–waveguide systems

• The appendix gives SLH representations for two-level atoms side-coupled to waveguides, including trapped atomic motion, guided and nonguided emission, recoil assumptions, and perfect-coupling conditions.The trapped atom includes position and momentum operators, propagation wave vectors, mass, and trap frequency; κ⊥ = 0 denotes perfect waveguide coupling.Source: Appendix A · p. 61

Atom–cavity systems

• Open atom–cavity systems are represented by Rabi and Jaynes–Cummings SLH triples, with the Tavis–Cummings model obtained by generalization.The rotating-wave approximation converts the Rabi model to the Jaynes–Cummings model, parameterized by cavity transmission, detunings, and atom–field coupling.Source: Appendix A · p. 62

Circulators

• Three- and four-port circulators are described by unitary scattering matrices, with transmission, reflection, and isolation-error coefficients characterizing nonideal behavior.For symmetric three-port circulators, reflection and isolation error are |r|^2 and |b|^2, while desirable operation requires |t| ≫ |r|, |b|.Source: Appendix A · p. 62