Introduction
Few-emitter laser theory usually models the gain medium as a two-level emitter with an incoherent pump term P D [ τ + ] P D[\tau_{+}] P D [ τ + ] [1 , 2 , 3 , 4 , 5 , 6 , 7 ] . This is an effective description. Real gain media pump through auxiliary levels: a Λ \Lambda Λ scheme empties the lower laser level into a reservoir that is pumped into the upper laser level, a four-level ladder separates the pump and laser transitions entirely, and a V V V scheme pumps the lower laser level into a fast-decaying upper reservoir. The distinction has consequences that the effective pump rate cannot capture. The pump acts as a dephasing channel for the lasing coherence only when it empties a laser level, so that the two-level model carries a pump-induced broadening that Λ \Lambda Λ and ladder media do not have; a pump cycle with several rate-limited steps delivers excitations more regularly than a Poisson process and can drive the output below the shot-noise level — the pump-noise suppression of regularly pumped lasers [8 , 9 ] in the dynamic form it takes in multilevel lasers [10 , 11 , 12 ] ; and the number of pump quanta an emitter must absorb before it can share a coherence with another emitter sets the order at which inter-emitter correlations appear. Single-ion lasers [13 ] , atoms in optical cavities [14 ] , and quantum-dot nanolasers [15 , 16 ] operate with such multilevel kinetics at the level of a few emitters and a few photons, where cumulant closures are least controlled [17 , 18 ] and exact finite-N N N solutions are the natural benchmark.
Exact permutation-invariant treatments of many multilevel emitters exist [19 , 20 , 21 , 22 , 23 ] . In particular, Gegg and Richter [22 ] showed that for incoherently pumped three- and four-level lasers the optical coherences that the dynamics never generates decouple from the permutation-invariant equations of motion, which reduces the scaling of the number of variables from N 8 N^8 N 8 to N 4 N^4 N 4 and from N 15 N^{15} N 15 to N 5 N^5 N 5 , and they solved the resulting Fock-space problems numerically. In a companion paper [24 ] , hereafter Paper I, we combined the Briegel–Englert damping basis of the cavity mode [25 , 26 ] with the permutation-invariant Liouville basis of N N N two-level emitters and showed that the coupled Liouvillian is an exact block-Jacobi operator in a single radial field index, whose stationary state is a matrix continued fraction with a linear, uniquely solvable asymptotic closure, and whose convergence to the stationary state of the infinite-dimensional master equation could be proved for a zero-temperature cavity under explicit hypotheses. The present paper carries that construction over to N N N identical d d d -level emitters with arbitrary independent incoherent level kinetics and one cavity-coupled transition, and uses it to answer the physical questions above. Methodologically it contributes a formulation of the reduced operator space for arbitrary incoherent rate graphs, generalizing the three- and four-level cases of Ref. [22 ] [Proposition 1], a single lifting rule that replaces scheme-specific multiplication tables (Lemma 1), a transfer of the structural and convergence results of Paper I with the hypotheses under which they hold (Proposition 2), a self-contained assembly prescription for the recurrence (Appendix F) with a stand-alone reference implementation, and a normalized-coordinate implementation that improves the conditioning of the recursion (Appendix E). Spectra and intensity correlations can be obtained from Fock-space solvers by quantum regression as well; what the damping basis adds is that they, the moment functionals, and a weak-pump series that terminates exactly in the radial index at each order all follow from one recurrence with a controlled structure.
Whether an effective two-level pump “fails” depends on what is held fixed and what is fitted, because a two-level model with more free parameters can be tuned to match more observables. We ask what a single Markovian pump jump gets wrong when the cavity ( A , ν , Δ ) (A,\nu,\Delta) ( A , ν , Δ ) , the coupling g g g , the laser-transition decay γ \gamma γ , and N N N are those of the multilevel medium and only the pump rate is fitted, to the mean photon number. This is the replacement that effective-pump models make; fits with additional free parameters (an extra dephasing, a renormalized γ \gamma γ or g g g ) are not tested here. The physical results are the following. (i) We prove that, for a zero-temperature cavity, the photon number of any scheme whose pump empties a laser level tends to zero as the pump rate grows (Proposition 4) — a sufficient mechanism for self-quenching, not a classification — and identify the explicit limiting models to which the Λ \Lambda Λ and ladder cycles studied here saturate instead; the exact solution follows the two-level and V lasers through self-quenching and the Λ \Lambda Λ and ladder lasers through saturation, and shows that matching the absorbed pump current reduces but does not remove the differences between schemes. (ii) We prove a pump-quantum filtration (Proposition 3) by which the pair coherence of Λ \Lambda Λ and ladder media is of second order in the pump, evaluate its leading coefficient from a finite linear system that is exact for the model, and map the recycling-rate boundary at which this coefficient changes sign. (iii) We show, by a controlled comparison at matched photon number, that rate-limited steps in the pump cycle lower the long-time output-counting Fano factor of two and three emitters below one, and the intracavity Fano factor as well at photon numbers below one; at a mean occupation of one the counting Fano factor of two emitters is 0.74 while g ( 2 ) ( 0 ) = 0.97 g^{(2)}(0)=0.97 g ( 2 ) ( 0 ) = 0.97 , and the time-resolved intensity correlation shows why: the anticorrelation is concentrated at delays of about half a cavity lifetime. The fitted two-level model gets g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , the intracavity and counting Fano factors, and the weak-pump order and sign of the pair coherence wrong; a reference of independent single-emitter lasers at the same output quantifies what the common cavity changes; and an independent Fock-space calculation confirms the counting statistics to 10 − 12 10^{-12} 1 0 − 12 . (iv) We decompose the output flux exactly over the collective sectors and quantify the population of the structurally dark sectors that exist for every N N N when d ≥ 3 d \ge3 d ≥ 3 . (v) We benchmark the second-order cumulant closure for multilevel media to N = 8 N=8 N = 8 . Neither pump-noise suppression nor exact permutation-invariant multilevel calculations are new; what is new here is their combination: several correlated emitters treated exactly with their full pump cycle, compared at matched intensity and against an independent-source reference, together with analytic selection rules for the pump order of the inter-emitter coherence and an exact-series determination of its recycling-dependent sign, and the demonstration that the temporal noise suppression survives where the equal-time statistics obscure it.
Section II fixes the model, Sec. III gives the exact solution in compact form, Secs. IV–VIII contain the physical results, and Sec. IX the scope and limitations. Appendix A contains the proofs and construction details of Sec. III, Appendix B records which properties of the emitters each inherited proof uses, Appendix C gives the multilevel closure equations, Appendix D the validation, cost, and conditioning, Appendix E the solver used for the largest emitter numbers, and Appendix F the explicit assembly of the recurrence. All code and data are available [citep][27].
Model and pump schemes
Each of N N N identical emitters has levels ∣ 1 ⟩ , … , ∣ d ⟩ \lvert1\rangle,\ldots,\lvert d\rangle ∣ 1 ⟩ , … , ∣ d ⟩ with projectors P i = ∣ i ⟩ ⟨ i ∣ P_i=\lvert i\rangle\langle i\rvert P i = ∣ i ⟩ ⟨ i ∣ . Two of them, the upper and lower laser levels u u u and l l l , are coupled to one cavity mode through τ + = ∣ u ⟩ ⟨ l ∣ \tau_+=\lvert u\rangle\langle l\rvert τ + = ∣ u ⟩ ⟨ l ∣ , τ − = τ + † \tau_-=\tau_+^\dagger τ − = τ + † , and the collective operators S ± = ∑ j τ ± ( j ) S_{\pm}=\sum_j\tau_{\pm}^{(j)} S ± = ∑ j τ ± ( j ) . Incoherent transitions i → i ′ i\to i' i → i ′ occur at rates γ i ′ ← i ≥ 0 \gamma_{i'\leftarrow i}\ge0 γ i ′ ← i ≥ 0 , and the lasing coherence may suffer pure dephasing at rate γ ϕ \gamma_\phi γ ϕ . With Γ i = ∑ i ′ ≠ i γ i ′ ← i \Gamma_i=\sum_{i'\ne i}\gamma_{i'\leftarrow i} Γ i = ∑ i ′ = i γ i ′ ← i the escape rate from level i i i , n ^ = a † a \hat n=a^\dagger a n ^ = a † a , and D [ L ] ρ = L ρ L † − 1 2 { L † L , ρ } \mathcal{D}[L]\rho=L\rho L^\dagger-\frac{1}{2}\{L^\dagger L,\rho\} D [ L ] ρ = L ρ L † − 2 1 { L † L , ρ } , the master equation is
ρ ˙ = − i ℏ [ H , ρ ] + A ( 1 + ν ) D [ a ] ρ + A ν D [ a † ] ρ + ∑ j = 1 N [ ∑ i ≠ i ′ γ i ′ ← i D [ ∣ i ′ ⟩ ⟨ i ∣ ( j ) ] ρ + 2 γ ϕ D [ P u ( j ) ] ρ ] \dot{\rho}=-\frac{i}{\hbar}[H,\rho]+A(1+\nu)\mathcal{D}[a]\rho+A\nu\mathcal{D}[a^\dagger]\rho
+\sum_{j=1}^{N}\left[\sum_{i\ne i'}\gamma_{i'\leftarrow i}\mathcal{D}[\lvert i'\rangle\langle i\rvert^{(j)}]\rho+2\gamma_\phi\mathcal{D}[P_u^{(j)}]\rho\right] ρ ˙ = − ℏ i [ H , ρ ] + A ( 1 + ν ) D [ a ] ρ + A ν D [ a † ] ρ + j = 1 ∑ N i = i ′ ∑ γ i ′ ← i D [∣ i ′ ⟩ ⟨ i ∣ ( j ) ] ρ + 2 γ ϕ D [ P u ( j ) ] ρ
H = ℏ ω n ^ + ℏ ∑ j ∑ i ϵ i P i ( j ) − ℏ g 2 ( a † S − + a S + ) , (1) H=\hbar\omega\hat n+\hbar\sum_j\sum_i\epsilon_iP_i^{(j)}-\frac{\hbar g}{2}\left(a^\dagger S_-+aS_+\right), \tag*{(1)} H = ℏ ω n ^ + ℏ j ∑ i ∑ ϵ i P i ( j ) − 2 ℏ g ( a † S − + a S + ) , (1)
with cavity decay rate A A A , thermal occupation ν \nu ν , transition frequency Ω = ϵ u − ϵ l \Omega=\epsilon_u-\epsilon_l Ω = ϵ u − ϵ l , and detuning Δ = Ω − ω \Delta=\Omega-\omega Δ = Ω − ω . The two-level model of Paper I is the case d = 2 d=2 d = 2 , γ l ← u = B ( 1 − s ) \gamma_{l\leftarrow u}=B(1-s) γ l ← u = B ( 1 − s ) , γ u ← l = B s \gamma_{u\leftarrow l}=Bs γ u ← l = B s , γ ϕ = C − B / 2 \gamma_\phi=C-B/2 γ ϕ = C − B /2 . Throughout, A > 0 A>0 A > 0 , g ≠ 0 g\ne0 g = 0 , and the rate graph is irreducible on the d d d levels, so that the classical generator
W i ′ i = γ i ′ ← i ( i ′ ≠ i ) , W i i = − Γ i , (2) W_{i'i}=\gamma_{i'\leftarrow i}\quad(i'\ne i),\qquad W_{ii}=-\Gamma_i, \tag*{(2)} W i ′ i = γ i ′ ← i ( i ′ = i ) , W ii = − Γ i , (2)
has a unique stationary distribution p ˉ \bar p p ˉ . Every Fock-truncated Lindbladian is then irreducible by the argument of Paper I and has a unique stationary state [citep][28, 29]. For the untruncated equation at ν = 0 \nu=0 ν = 0 we keep the moment-stationarity uniqueness hypothesis (U) of Paper I with the excitation operator E = n ^ + N u \mathcal{E}=\hat n+N_u E = n ^ + N u , N u = ∑ j P u ( j ) N_u=\sum_jP_u^{(j)} N u = ∑ j P u ( j ) : every trace-class σ \sigma σ with e θ E σ e θ E e^{\theta\mathcal{E}}\sigma e^{\theta\mathcal{E}} e θ E σ e θ E trace class for all θ > 0 \theta>0 θ > 0 and tr ( σ L † Y ) = 0 \operatorname{tr}(\sigma L^\dagger Y)=0 tr ( σ L † Y ) = 0 for all polynomials Y Y Y in the field and emitter operators is a multiple of one density operator ρ s s \rho_{\mathrm{ss}} ρ ss . The Lyapunov inequality that makes (U) plausible in Paper I holds with L † E ≤ − min ( A , Γ u ) E + Γ ↑ N + A ν \mathcal{L}^\dagger\mathcal{E}\le-\min(A,\Gamma_u)\mathcal{E}+\Gamma_\uparrow N+A\nu L † E ≤ − min ( A , Γ u ) E + Γ ↑ N + A ν , Γ ↑ = max i ≠ u γ u ← i \Gamma_\uparrow=\max_{i\ne u}\gamma_{u\leftarrow i} Γ ↑ = max i = u γ u ← i , provided Γ u > 0 \Gamma_u>0 Γ u > 0 .
Three schemes serve as examples throughout (Figure 1 , Table 1 ): the V scheme and the Λ \Lambda Λ scheme (d = 3 d=3 d = 3 ) and the four-level ladder (d = 4 d=4 d = 4 ). What distinguishes them is which level the pump empties. In the two-level and V schemes it empties the lower laser level; in the Λ \Lambda Λ scheme and the ladder it empties a reservoir, and neither laser level is emptied by the pump — the lower laser level is emptied by recycling instead. All rates are quoted in units of the cavity decay rate A A A . Two parameter regimes are used: the good-cavity regime of Paper I, g = 5 A g = 5A g = 5 A , γ = A \gamma= A γ = A (its decay rate B ( 1 − s ) B(1-s) B ( 1 − s ) at B = 10 A B = 10A B = 10 A , s = 0.9 s = 0.9 s = 0.9 ; the two-level laser at P = 9 A P = 9A P = 9 A is then the good-cavity laser of Paper I and reproduces its tbl:8 for N = 1 N = 1 N = 1 –3 3 3 ), in which the lasers of § 3 reach several photons, and the bad-cavity regime g = 1.1 A g = 1.1A g = 1.1 A , γ = 0.3 A \gamma= 0.3A γ = 0.3 A , in which the few-quanta correlation results of Secs. IV, VII, and VIII are obtained.
Figure 1. The four pump schemes. Solid blue: cavity-coupled laser transition (coupling g g g ); orange: incoherent pump (rate P P P ); dashed grey: incoherent decays. (a) Two-level emitter: the pump empties the lower laser level into the upper one. (b) V scheme: the lower laser level is pumped into a reservoir r r r that decays at rate γ r \gamma_r γ r into the upper laser level. (c) Λ \Lambda Λ scheme: a reservoir below the laser levels is pumped into the upper laser level; the lower laser level is emptied into the reservoir at the recycling rate γ r \gamma_r γ r . (d) Four-level ladder: ground → \to → pump level (P P P ) → \to → upper laser level (γ p \gamma_p γ p ) → \to → lower laser level (cavity and γ \gamma γ ) → \to → ground (γ r \gamma_r γ r ). In all schemes the laser transition also decays spontaneously at rate γ = γ l ← u \gamma= \gamma_{l\leftarrow u} γ = γ l ← u .
Table 1. The example schemes. C u l C_{\mathrm{ul}} C ul is the transverse damping rate of the lasing coherence, Eq. (4) , at γ ϕ = 0 \gamma_\phi=0 γ ϕ = 0 ; D N D_N D N the atomic block dimension (L = d + 2 L=d+2 L = d + 2 local operators, Proposition 1); χ \chi χ the pump-quantum charge of Proposition 3 that gives the sharpest weak-pump orders, with the resulting order of the pair coherence; and the P → ∞ P\to\infty P → ∞ limit of the scheme at ν = 0 \nu=0 ν = 0 (§ 3 ).
Exact solution
This section states the construction; proofs and the details of the local bases are in Appendix A, and the reader interested in the physics may skip to Sec. IV after Eq. (8) .
Reduced local operator space and collective basis
With one cavity-coupled transition and incoherent everything else, the dynamics of one emitter never leaves the ( d + 2 ) (d+2) ( d + 2 ) -dimensional space A = span { P 1 , … , P d , τ + , τ − } \mathcal{A} = \operatorname{span}\{P_1,\ldots,P_d,\tau_+,\tau_-\} A = span { P 1 , … , P d , τ + , τ − } : jumps map populations to populations and only damp the coherence they touch, and the cavity generates no coherence except ∣ u ⟩ ⟨ l ∣ |u\rangle\langle l| ∣ u ⟩ ⟨ l ∣ and its conjugate. For incoherently pumped three- and four-level lasers this decoupling of the unused coherences was demonstrated and exploited by Gegg and Richter [22 ] ; the following proposition states it for arbitrary independent incoherent rate graphs.
Proposition 1 (Invariant local operator space). (i) The one-emitter Liouvillian of Eq. (1), including the level Hamiltonian, maps A \mathcal{A} A into itself; (ii) left and right multiplication by τ ± \tau_\pm τ ± map A \mathcal{A} A into itself; (iii) 1 ∈ A \mathbb{1} \in\mathcal{A} 1 ∈ A . Consequently the span of the field damping-basis operators tensored with the permutation-invariant products of elements of A \mathcal{A} A is invariant under the full Liouvillian, and the stationary state of every Fock truncation, as well as ρ s s \rho_{\mathrm{ss}} ρ ss under (U), lies in it. The atomic block dimension is
D N ( d ) = ( N + d + 1 d + 1 ) , (3) D_N^{(d)} = \binom{N+d+1}{d+1}, \tag*{(3)} D N ( d ) = ( d + 1 N + d + 1 ) , (3)
( N + 4 4 ) for three and ( N + 5 5 ) for four levels, against ( N + d 2 − 1 d 2 − 1 ) for the full local Liouville space (Table V). \binom{N+4}{4}\ \text{for three and}\ \binom{N+5}{5}\ \text{for four levels, against}\ \binom{N+d^2-1}{d^2-1}\ \text{for the full local Liouville space (Table V).} ( 4 N + 4 ) for three and ( 5 N + 5 ) for four levels, against ( d 2 − 1 N + d 2 − 1 ) for the full local Liouville space (Table V).
A basis { b μ } \{b_\mu\} { b μ } of A \mathcal{A} A with exactly one traceful element b 0 b_0 b 0 , traceless b 1 , … , b d − 1 b_1,\ldots,b_{d-1} b 1 , … , b d − 1 , and b ± = τ ± b_\pm=\tau_\pm b ± = τ ± is chosen; we use the eigenoperators of the one-emitter Liouvillian (the local damping basis) where its rate matrix is diagonalizable and a rate-independent fixed basis otherwise (Appendix A). In either, τ ± \tau_\pm τ ± are eigenoperators with eigenvalues − C u l ∓ i Ω -C_{ul}\mp i\Omega − C u l ∓ i Ω ,
C u l = 1 2 ( Γ u + Γ l ) + γ ϕ , (4) C_{ul} = \frac{1}{2}(\Gamma_u+\Gamma_l)+\gamma_\phi, \tag*{(4)} C u l = 2 1 ( Γ u + Γ l ) + γ ϕ , (4)
the transverse rate of the lasing coherence: half the sum of the escape rates of the two laser levels plus pure dephasing. For the two-level model this is C = B / 2 + γ ϕ C = B/2+\gamma_\phi C = B /2 + γ ϕ ; for the V scheme C u l = ( γ + P ) / 2 C_{ul}=(\gamma+P)/2 C u l = ( γ + P ) /2 grows with the pump exactly as in the two-level model; for the Λ \Lambda Λ scheme and the ladder C u l = ( γ + γ r ) / 2 C_{ul}=(\gamma+\gamma_r)/2 C u l = ( γ + γ r ) /2 is pump independent. The cavity is expanded in the damping basis of Briegel and Englert [25 ] , the right eigenoperators ρ n ( k ) \rho_n^{(k)} ρ n ( k ) of the field Liouvillian with radial index n ≥ 0 n\geq0 n ≥ 0 , coherence grade k ∈ Z k\in\mathbb{Z} k ∈ Z , eigenvalues − A ( n + ∣ k ∣ / 2 ) − i ω k -A(n+|k|/2)-i\omega k − A ( n + ∣ k ∣/2 ) − iω k , and tr ρ n ( k ) = δ n 0 δ k 0 \operatorname{tr}\rho_n^{(k)}=\delta_{n0}\delta_{k0} tr ρ n ( k ) = δ n 0 δ k 0 . The permutation-invariant atomic basis consists of the averaged symmetrized products
R m = 1 N ( m ) ∑ π d i s t i n c t π [ ⨂ μ b μ ⊗ m μ ] , N ( m ) = N ! ∏ μ m μ ! , (5) R_m = \frac{1}{\mathcal{N}(m)}\sum_{\pi}^{\mathrm{distinct}} \pi\left[\bigotimes_\mu b_\mu^{\otimes m_\mu}\right],\qquad\mathcal{N}(m)=\frac{N!}{\prod_\mu m_\mu!}, \tag*{(5)} R m = N ( m ) 1 π ∑ distinct π [ μ ⨂ b μ ⊗ m μ ] , N ( m ) = ∏ μ m μ ! N ! , (5)
with m = ( m 0 , … , m d − 1 , m + , m − ) m=(m_0,\ldots,m_{d-1},m_+,m_-) m = ( m 0 , … , m d − 1 , m + , m − ) , ∣ m ∣ = N |\mathbf{m}|=N ∣ m ∣ = N , tr R m = δ m , N e 0 \operatorname{tr}R_m=\delta_{m,N e_0} tr R m = δ m , N e 0 , and coherence charge q ( m ) = m + − m − q(m)=m_+-m_- q ( m ) = m + − m − . The action of any one-site map on this basis follows from one rule, which replaces the four multiplication tables of Paper I.
Lemma 1 (Lifting rule). Let x x x be a linear map on A \mathcal{A} A with structure constants x b μ = ∑ ν x ν μ b ν x b_\mu=\sum_\nu x_{\nu\mu}b_\nu x b μ = ∑ ν x ν μ b ν , and x ( N ) = ∑ j x ( j ) x^{(N)}=\sum_j x^{(j)} x ( N ) = ∑ j x ( j ) . Then
x ( N ) R m = ∑ μ m μ ∑ ν x ν μ R m − e μ + e ν . (6) x^{(N)}R_m=\sum_\mu m_\mu\sum_\nu x_{\nu\mu}R_{m-e_\mu+e_\nu}. \tag*{(6)} x ( N ) R m = μ ∑ m μ ν ∑ x ν μ R m − e μ + e ν . (6)
Block-Jacobi recurrence and what carries over
The total coherence K = k + q ( m ) K=k+q(m) K = k + q ( m ) of a product ρ n ( k ) ⊗ R m \rho_n^{(k)}\otimes R_m ρ n ( k ) ⊗ R m is conserved by the interaction. With an ordering { m α } \{m_\alpha\} { m α } , k α = K − q α k_\alpha=K-q_\alpha k α = K − q α , Φ n α ( K ) = ρ n ( k α ) ⊗ R m α \Phi_{n\alpha}^{(K)}=\rho_n^{(k_\alpha)}\otimes R_{m_\alpha} Φ n α ( K ) = ρ n ( k α ) ⊗ R m α , and ρ = ∑ n , α ( X n ) α Φ n α ( K ) \rho=\sum_{n,\alpha}(X_n)_\alpha\Phi_{n\alpha}^{(K)} ρ = ∑ n , α ( X n ) α Φ n α ( K ) , the lifted matrices of left and right multiplication by τ ± \tau_\pm τ ± and the field coefficients of the damping basis give blocks M n , G n , F n ∈ C D N × D N M_n,G_n,F_n\in\mathbb{C}^{D_N\times D_N} M n , G n , F n ∈ C D N × D N , (18) of Appendix F, which lists every ingredient needed to implement them, with the diagonal uncoupled part − A ( n + ∣ k β ∣ / 2 ) + λ β a t − i ω K − i Δ q β -A(n+|k_\beta|/2)+\lambda_\beta^{\mathrm{at}}-i\omega K-i\Delta q_\beta − A ( n + ∣ k β ∣/2 ) + λ β at − iω K − i Δ q β , where λ β a t = − ∑ κ m κ β β κ − C u l ( m + β + m − β ) \lambda_\beta^{\mathrm{at}}=-\sum_\kappa m_{\kappa\beta}\beta_\kappa-C_{\mathrm{ul}}(m_{+\beta}+m_{-\beta}) λ β at = − ∑ κ m κ β β κ − C ul ( m + β + m − β ) in the damping basis (in the fixed basis the atomic matrix is not diagonal, which nothing below requires). Call ∣ k β ∣ |k_\beta| ∣ k β ∣ the grade of a basis state; it ranges over 0 , … , N + ∣ K ∣ 0,\ldots,N+|K| 0 , … , N + ∣ K ∣ , because ∣ q ∣ ≤ N |q|\leq N ∣ q ∣ ≤ N for one cavity-coupled transition.
Proposition 2 (Structure of the d d d -level recurrence). For N N N identical d d d -level emitters obeying Eq. (1) and any fixed K K K : (i) the Liouvillian on the permutation-invariant sector is exactly block tridiagonal in the radial index, X ˙ n = M n X n + G n X n + 1 + F n X n − 1 \dot{X}_n=M_nX_n+G_nX_{n+1}+F_nX_{n-1} X ˙ n = M n X n + G n X n + 1 + F n X n − 1 , X − 1 = 0 X_{-1}=0 X − 1 = 0 ; (ii) G n = G G_n=G G n = G is constant, F n = n F ( 1 ) F_n=nF^{(1)} F n = n F ( 1 ) , M n = n M ( 1 ) + M ( 0 ) M_n=nM^{(1)}+M^{(0)} M n = n M ( 1 ) + M ( 0 ) with M ( 1 ) = − A 1 + i g 2 N K M^{(1)}=-A\mathbb{1}+\frac{ig}{2}N_K M ( 1 ) = − A 1 + 2 i g N K , N K = [ S + , ⋅ ] P q < K + [ S − , ⋅ ] P q > K N_K=[S_+,\cdot]P_{q<K}+[S_-,\cdot]P_{q>K} N K = [ S + , ⋅ ] P q < K + [ S − , ⋅ ] P q > K , so that a finite limit of the transfer matrices X n + 1 = R n X n X_{n+1}=R_nX_n X n + 1 = R n X n satisfies the linear equation M ( 1 ) R ∞ = − F ( 1 ) M^{(1)}R_\infty=-F^{(1)} M ( 1 ) R ∞ = − F ( 1 ) , uniquely solvable for every A > 0 A>0 A > 0 by the finite Neumann series
R ∞ = A − 1 ∑ j = 0 N + ∣ K ∣ ( i g 2 A ) j N K j F ( 1 ) ; R_\infty=A^{-1}\sum_{j=0}^{N+|K|}\left(\frac{ig}{2A}\right)^jN_K^jF^{(1)}; R ∞ = A − 1 j = 0 ∑ N + ∣ K ∣ ( 2 A i g ) j N K j F ( 1 ) ; (iii) F ( 1 ) F^{(1)} F ( 1 ) and N K N_K N K lower the grade of every basis state by exactly one, G G G raises it by exactly one, and R ∞ R_\infty R ∞ lowers it by at least one; all four are nilpotent of index at most N + ∣ K ∣ + 1 N+|K|+1 N + ∣ K ∣ + 1 ; (iv) for every terminal matrix R n max R_{n_{\max}} R n m a x in a ball ∥ R ∥ ≤ ρ 0 \|R\|\leq\rho_0 ∥ R ∥ ≤ ρ 0 , ρ 0 > ∥ R ∞ ∥ \rho_0>\|R_\infty\| ρ 0 > ∥ R ∞ ∥ , the backward recursion R n = − ( M n + 1 + G R n + 1 ) − 1 F n + 1 R_n=-(M_{n+1}+GR_{n+1})^{-1}F_{n+1} R n = − ( M n + 1 + G R n + 1 ) − 1 F n + 1 converges as n max → ∞ n_{\max}\to\infty n m a x → ∞ , for all n ≥ n 0 ( ρ 0 ) n\geq n_0(\rho_0) n ≥ n 0 ( ρ 0 ) , to limits R n = R ∞ + O ( 1 / n ) R_n=R_\infty+O(1/n) R n = R ∞ + O ( 1/ n ) independent of the terminal matrix; the same holds for all n ≥ 0 n\geq0 n ≥ 0 under (P) the limiting pivots M j + G R ∞ M_j+GR_\infty M j + G R ∞ are invertible for 1 ≤ j ≤ n 0 1\leq j\leq n_0 1 ≤ j ≤ n 0 (for j > n 0 j>n_0 j > n 0 this is automatic); (v) at ν = 0 \nu=0 ν = 0 the stationary state of every Fock truncation obeys ⟨ e θ ⟩ ≤ 2 e θ ( N + n ∗ ) \langle e^\theta\rangle\leq2e^{\theta(N+n_\ast)} ⟨ e θ ⟩ ≤ 2 e θ ( N + n ∗ ) , n ∗ = ⌈ 2 Γ ↑ N / ( A e − θ ) ⌉ n_\ast=\lceil2\Gamma_\uparrow N/(Ae^{-\theta})\rceil n ∗ = ⌈ 2 Γ ↑ N / ( A e − θ )⌉ , uniformly in the cutoff, the Fock truncations converge in trace norm to ρ s s \rho_{\mathrm{ss}} ρ ss under (U), and for K = 0 K=0 K = 0 , under (U) and (P), the normalized solution of the continued fraction converges componentwise to the damping-basis coefficients for ρ s s \rho_{\mathrm{ss}} ρ ss and every observable functional below to its stationary value.
Items (i)–(iv) are Propositions 1–4 of Paper I and item (v) is its Lemma 1, Proposition 5, and Theorem 1; Appendix B lists, item by item, the only properties of the emitter sector that the original proofs use—essentially that τ + \tau_+ τ + raises the coherence charge by exactly one on every nonvanishing term of Eq. (6) , that ∣ q ∣ ≤ N |q|\leq N ∣ q ∣ ≤ N , and that the jumps change E \mathcal{E} E by 0 or ± 1 \pm1 ± 1 with bounded raising rate— and verifies that each holds here. As in Paper I, (P) concerns the finitely many non-asymptotic limiting pivots and remains an assumption whose finite-cutoff proxies are monitored; and the regression spectrum of the K = − 1 K=-1 K = − 1 sector and the intensity correlation of the K = 0 K=0 K = 0 sector are finite-truncation eigenvalue problems whose convergence is checked numerically, not covered by (v).
The emission spectrum follows from the K = − 1 K=-1 K = − 1 sector by quantum regression as in Sec. IV E of Paper I, and the intensity correlation from the K = 0 K=0 K = 0 sector (Sec. VI); the numerical solution of the recurrence is described in Appendices A and E.
Observables and photon balance
Because b 0 b_0 b 0 is the only traceful local element and tr ρ n ( k ) = δ n 0 δ k 0 \operatorname{tr}\rho_n^{(k)}=\delta_{n0}\delta_{k0} tr ρ n ( k ) = δ n 0 δ k 0 : with c n c_n c n the component of X n X_n X n along N e 0 Ne_0 N e 0 , ⟨ n ^ ⟩ = ν + ( 1 + ν ) c 1 \langle\hat{n}\rangle=\nu+(1+\nu)c_1 ⟨ n ^ ⟩ = ν + ( 1 + ν ) c 1 and ⟨ n ^ ( n ^ − 1 ) ⟩ = 2 ν 2 + 4 ν ( 1 + ν ) c 1 + 2 ( 1 + ν ) 2 c 2 \langle\hat{n}(\hat{n}-1)\rangle=2\nu^2+4\nu(1+\nu)c_1+2(1+\nu)^2c_2 ⟨ n ^ ( n ^ − 1 )⟩ = 2 ν 2 + 4 ν ( 1 + ν ) c 1 + 2 ( 1 + ν ) 2 c 2 ; the reduced atomic state is ρ N = ∑ q α = 0 ( X 0 ) α 0 R m α \rho_N=\sum_{q_\alpha=0}(X_0)_{\alpha0}R_{m_\alpha} ρ N = ∑ q α = 0 ( X 0 ) α 0 R m α ; for a one-emitter observable O O O and a two-emitter product O O ′ OO' O O ′ , with ρ μ = tr ( b μ O ) \rho_\mu=\operatorname{tr}(b_\mu O) ρ μ = tr ( b μ O ) , ρ μ ′ = tr ( b μ O ′ ) \rho'_\mu=\operatorname{tr}(b_\mu O') ρ μ ′ = tr ( b μ O ′ ) ,
⟨ ∑ j O ( j ) ⟩ = N O 0 X 0 [ N e 0 ] + ∑ μ ≠ 0 ρ μ X 0 [ ( N − 1 ) e 0 + e μ ] , \left\langle\sum_j O^{(j)}\right\rangle= N O_0X_0[Ne_0]+\sum_{\mu\ne0}\rho_\mu X_0[(N-1)e_0+e_\mu], ⟨ j ∑ O ( j ) ⟩ = N O 0 X 0 [ N e 0 ] + μ = 0 ∑ ρ μ X 0 [( N − 1 ) e 0 + e μ ] ,
⟨ ∑ i ≠ j O ( i ) O ′ ( j ) ⟩ = N ( N − 1 ) ρ 0 ρ 0 ′ X 0 [ N e 0 ] + ( N − 1 ) ∑ μ ≠ 0 ( ρ μ ρ 0 ′ + ρ 0 ρ μ ′ ) X 0 [ ( N − 1 ) e 0 + e μ ] + ∑ μ , ν ≠ 0 ( 1 + δ μ ν ) ρ μ ρ ν ′ X 0 [ ( N − 2 ) e 0 + e μ + e ν ] . (7) \left\langle\sum_{i\ne j}O^{(i)}O'^{(j)}\right\rangle=N(N-1)\rho_0\rho'_0X_0[Ne_0]
+(N-1)\sum_{\mu\ne0}(\rho_\mu\rho'_0+\rho_0\rho'_\mu)X_0[(N-1)e_0+e_\mu]
+\sum_{\mu,\nu\ne0}(1+\delta_{\mu\nu})\rho_\mu\rho'_\nu X_0[(N-2)e_0+e_\mu+e_\nu]. \tag*{(7)} ⟨ i = j ∑ O ( i ) O ′ ( j ) ⟩ = N ( N − 1 ) ρ 0 ρ 0 ′ X 0 [ N e 0 ] + ( N − 1 ) μ = 0 ∑ ( ρ μ ρ 0 ′ + ρ 0 ρ μ ′ ) X 0 [( N − 1 ) e 0 + e μ ] + μ , ν = 0 ∑ ( 1 + δ μν ) ρ μ ρ ν ′ X 0 [( N − 2 ) e 0 + e μ + e ν ] . (7)
so that the pair coherence is C 2 ( N ) = ⟨ τ + ( 1 ) τ − ( 2 ) ⟩ = X 0 [ ( N − 2 ) e 0 + e + + e − ] / [ N ( N − 1 ) ] C_2^{(N)}=\langle\tau_+^{(1)}\tau_-^{(2)}\rangle=X_0[(N-2)e_0+e_++e_-]/[N(N-1)] C 2 ( N ) = ⟨ τ + ( 1 ) τ − ( 2 ) ⟩ = X 0 [( N − 2 ) e 0 + e + + e − ] / [ N ( N − 1 )] . Joint observables ⟨ a † O ⟩ \langle a^\dagger O\rangle ⟨ a † O ⟩ with O O O atomic of charge − 1 -1 − 1 are ∑ q α = + 1 ( X 0 ) α tr ( R m α O ) \sum_{q_\alpha=+1}(X_0)_\alpha\operatorname{tr}(R_{m_\alpha}O) ∑ q α = + 1 ( X 0 ) α tr ( R m α O ) .
The exact photon balance of Paper I generalizes with the laser-level inversion W u l = ∑ j ( P u ( j ) − P l ( j ) ) W_{\mathrm{ul}}=\sum_j(P_u^{(j)}-P_l^{(j)}) W ul = ∑ j ( P u ( j ) − P l ( j ) ) and C u l C_{\mathrm{ul}} C ul : with Z = ⟨ a † S − ⟩ Z=\langle a^\dagger S_-\rangle Z = ⟨ a † S − ⟩ ,
A ( ⟨ n ^ ⟩ − ν ) = ∑ i ≠ u γ u ← i ⟨ N i ⟩ − Γ u ⟨ N u ⟩ = − g Im Z , A(\langle\hat{n}\rangle-\nu)=\sum_{i\ne u}\gamma_{u\leftarrow i}\langle N_i\rangle-\Gamma_u\langle N_u\rangle=-g\operatorname{Im}Z, A (⟨ n ^ ⟩ − ν ) = i = u ∑ γ u ← i ⟨ N i ⟩ − Γ u ⟨ N u ⟩ = − g Im Z ,
A ( ⟨ n ^ ⟩ − ν ) = R u l [ ⟨ S + S − ⟩ + ⟨ n ^ W u l ⟩ ] , A(\langle\hat{n}\rangle-\nu)=R_{ul}\left[\langle S_+S_-\rangle+\langle\hat{n}W_{ul}\rangle\right], A (⟨ n ^ ⟩ − ν ) = R u l [ ⟨ S + S − ⟩ + ⟨ n ^ W u l ⟩ ] ,
R u l = g 2 ( A / 2 + C u l ) 2 [ ( A / 2 + C u l ) 2 + Δ 2 ] . (8) R_{ul}=\frac{g^2(A/2+C_{ul})}{2\left[(A/2+C_{ul})^2+\Delta^2\right]}. \tag*{(8)} R u l = 2 [ ( A /2 + C u l ) 2 + Δ 2 ] g 2 ( A /2 + C u l ) . (8)
where the first line is the stationary balance of the upper-laser-level population and the second follows from Z ˙ = − ( A / 2 + C u l + i Δ ) Z − i g 2 [ ⟨ S + S − ⟩ + ⟨ n ^ W u l ⟩ ] \dot{Z}=-(A/2+C_{ul}+i\Delta)Z-\frac{ig}{2}\left[\langle S_+S_-\rangle+\langle\hat{n}W_{ul}\rangle\right] Z ˙ = − ( A /2 + C u l + i Δ ) Z − 2 i g [ ⟨ S + S − ⟩ + ⟨ n ^ W u l ⟩ ] , which uses [ τ − , τ + ] = P l − P u [\tau_-,\tau_+]=P_l-P_u [ τ − , τ + ] = P l − P u and the eigenvalue of τ − \tau_- τ − under the one-emitter Liouvillian. For N = 2 N=2 N = 2 , ⟨ S + S − ⟩ = ⟨ N u ⟩ + 2 C 12 \langle S_+S_-\rangle=\langle N_u\rangle+2C_{12} ⟨ S + S − ⟩ = ⟨ N u ⟩ + 2 C 12 and the singlet–triplet representation 2 C 12 = p 1 , 0 − p 0 , 0 2C_{12}=p_{1,0}-p_{0,0} 2 C 12 = p 1 , 0 − p 0 , 0 of Paper I holds unchanged, the two states being the symmetric and antisymmetric combinations of ∣ u l ⟩ |ul\rangle ∣ u l ⟩ , ∣ l u ⟩ |lu\rangle ∣ l u ⟩ : the sign of the pair coherence is the sign of the population difference between the bright and the dark pair state within the subspace in which both emitters occupy laser levels.
Weak pump: pump-quantum filtration
Paper I proved, from a filtration of the Liouville space by the total excitation, that for two-level emitters the joint cumulants K m = κ ( n 1 , … , n m ) K_m=\kappa(n_1,\ldots,n_m) K m = κ ( n 1 , … , n m ) of the excitation numbers n j = P u ( j ) n_j=P_u^{(j)} n j = P u ( j ) are O ( P m ) O(P^m) O ( P m ) and the connected coherences Γ m = κ ( τ + ( 1 ) , τ − ( 2 ) , n 3 , … , n m ) \Gamma_m=\kappa(\tau_+^{(1)},\tau_-^{(2)},n_3,\ldots,n_m) Γ m = κ ( τ + ( 1 ) , τ − ( 2 ) , n 3 , … , n m ) are O ( P m − 1 ) O(P^{m-1}) O ( P m − 1 ) as P → 0 P\to0 P → 0 (Γ 2 = C 2 \Gamma_2=C_2 Γ 2 = C 2 ). For multilevel schemes the right grading is not the excitation but the number of pump quanta absorbed. Write L = L 0 + P L 1 \mathcal{L}=\mathcal{L}_0+P\mathcal{L}_1 L = L 0 + P L 1 with L 1 \mathcal{L}_1 L 1 the pump dissipators, and suppose that at P = 0 P=0 P = 0 the unique stationary state is ∣ 0 ⟩ ⟨ 0 ∣ ⊗ ∣ g ⋯ g ⟩ ⟨ g ⋯ g ∣ |0\rangle\langle0|\otimes|g\cdots g\rangle\langle g\cdots g| ∣0 ⟩ ⟨ 0∣ ⊗ ∣ g ⋯ g ⟩ ⟨ g ⋯ g ∣ for one level g g g . Call a function χ \chi χ from the levels to the nonnegative integers with χ ( g ) = 0 \chi(g)=0 χ ( g ) = 0 an admissible charge if every pump jump raises χ \chi χ by exactly one, no jump of L 0 \mathcal{L}_0 L 0 raises it, and either (a) χ ( u ) = χ ( l ) + 1 \chi(u)=\chi(l)+1 χ ( u ) = χ ( l ) + 1 , with E χ = n ^ + ∑ j χ ( j ) \mathcal{E}_\chi=\hat{n}+\sum_j\chi^{(j)} E χ = n ^ + ∑ j χ ( j ) , or (b) χ ( u ) = χ ( l ) \chi(u)=\chi(l) χ ( u ) = χ ( l ) , with E χ = ∑ j χ ( j ) \mathcal{E}_\chi=\sum_j\chi^{(j)} E χ = ∑ j χ ( j ) . In both cases H H H commutes with E χ \mathcal{E}_\chi E χ , the jumps of L 0 \mathcal{L}_0 L 0 (including (a)) do not raise it, and the pump jumps raise it by one. Let V r = span { ∣ i ⟩ ⟨ j ∣ : E i ≤ r , E j ≤ r } V_r=\operatorname{span}\{|i\rangle\langle j|:E_i\le r,E_j\le r\} V r = span { ∣ i ⟩ ⟨ j ∣ : E i ≤ r , E j ≤ r } in the eigenbasis of E χ \mathcal{E}_\chi E χ , and consider, in any Fock truncation, the regular expansion ρ s s ( P ) = ∑ r P r ρ ( r ) \rho_{\mathrm{ss}}(P)=\sum_rP^r\rho^{(r)} ρ ss ( P ) = ∑ r P r ρ ( r ) , which exists because zero is a simple eigenvalue of the truncated L 0 \mathcal{L}_0 L 0 .
Proposition 3 (Pump-quantum filtration, ν = 0 \nu=0 ν = 0 ). For every admissible charge χ \chi χ : L 0 V r ⊆ V r \mathcal{L}_0V_r\subseteq V_r L 0 V r ⊆ V r , L 1 V r ⊆ V r + 1 \mathcal{L}_1V_r\subseteq V_{r+1} L 1 V r ⊆ V r + 1 , and ρ ( r ) ∈ V r \rho^{(r)}\in V_r ρ ( r ) ∈ V r . Consequently every operator O = Π ≥ k O Π ≥ k O=\Pi_{\ge k}O\Pi_{\ge k} O = Π ≥ k O Π ≥ k supported on E χ ≥ k \mathcal{E}_\chi\ge k E χ ≥ k has ⟨ O ⟩ = O ( P k ) \langle O\rangle=O(P^k) ⟨ O ⟩ = O ( P k ) at the level of the series,
K m = O ( P m χ u ) , C 2 ( N ) = O ( P χ u + χ l ) , K_m=O(P^{m\chi_u}),\qquad C_2^{(N)}=O(P^{\chi_u+\chi_l}), K m = O ( P m χ u ) , C 2 ( N ) = O ( P χ u + χ l ) , Γ m = O ( P χ u + χ l + ( m − 2 ) χ u ) . (9) \Gamma_m=O(P^{\chi_u+\chi_l+(m-2)\chi_u}). \tag*{(9)} Γ m = O ( P χ u + χ l + ( m − 2 ) χ u ) . (9) The Taylor coefficients ρ ( r ) \rho^{(r)} ρ ( r ) are the same for every Fock cutoff n c > r n_c>r n c > r whenever a charge of type (a) is admissible, because V r V_r V r then contains no Fock level above $r; different admissible charges may be used simultaneously.
Proof. The proof of Lemma 2 of Paper I uses only that H H H commutes with the grading operator, that every jump of L 0 \mathcal{L}_0 L 0 lowers or preserves it, that every pump jump raises it by at most one, that ρ ( 0 ) ∈ V 0 \rho^{(0)}\in V_0 ρ ( 0 ) ∈ V 0 , and that zero is a simple eigenvalue of L 0 \mathcal{L}_0 L 0 on each finite-dimensional invariant subspace V r V_r V r , which follows from the uniqueness of the stationary state of the truncated L 0 \mathcal{L}_0 L 0 . All hold for E χ \mathcal{E}_\chi E χ . The monomial n i 1 ⋯ n i k n_{i_1}\cdots n_{i_k} n i 1 ⋯ n i k is supported on E χ ≥ k χ u \mathcal{E}_\chi\ge k\chi_u E χ ≥ k χ u . The operator τ + ( 1 ) τ − ( 2 ) n j 1 ⋯ n j b \tau_+^{(1)}\tau_-^{(2)}n_{j_1}\cdots n_{j_b} τ + ( 1 ) τ − ( 2 ) n j 1 ⋯ n j b maps states with emitter 1 in l l l , emitter 2 in u u u , and b b b further emitters in u u u to states with emitter 1 in u u u , emitter 2 in l l l , and the same b b b emitters in u u u ; both have E χ ≥ χ u + χ l + b χ u \mathcal{E}_\chi\ge\chi_u+\chi_l+b\chi_u E χ ≥ χ u + χ l + b χ u . The partition expansion of the cumulants then gives Eq. (9) term by term, as in Propositions 6 and 7 of Paper I. The last statement holds because E χ \mathcal{E}_\chi E χ of type (a) bounds n ^ \hat{n} n ^ by r r r . □
The admissible charges of the four schemes are listed in Table 1 . For the two-level and V schemes the lower laser level is the ground state, only charges of type (a) give nontrivial orders, and Eq. (9) reproduces the orders of Paper I: C 2 = O ( P ) C_2=O(P) C 2 = O ( P ) , Γ 3 = O ( P 2 ) \Gamma_3=O(P^2) Γ 3 = O ( P 2 ) , K 3 = O ( P 3 ) K_3=O(P^3) K 3 = O ( P 3 ) . For the Λ \Lambda Λ scheme and the ladder the lower laser level is itself one pump quantum above the ground state, a charge of type (b) with χ u = χ l = 1 \chi_u=\chi_l=1 χ u = χ l = 1 is admissible, and the pair coherence is second order in the pump, C 2 = O ( P 2 ) C_2=O(P^2) C 2 = O ( P 2 ) , Γ 3 = O ( P 3 ) \Gamma_3=O(P^3) Γ 3 = O ( P 3 ) , K 3 = O ( P 3 ) K_3=O(P^3) K 3 = O ( P 3 ) : two emitters must each absorb a pump quantum before they can share a coherence in the laser manifold. These two schemes also admit type-(a) charges — χ r = χ l = 0 , χ u = 1 \chi_r=\chi_l=0,\ \chi_u=1 χ r = χ l = 0 , χ u = 1 for Λ \Lambda Λ , and χ g = χ l = 0 , χ p = χ u = 1 \chi_g=\chi_l=0,\ \chi_p=\chi_u=1 χ g = χ l = 0 , χ p = χ u = 1 for the ladder — which do not give the sharpest orders but bound the photon number on V r V_r V r , so that their Taylor coefficients are cutoff independent as well. (The direct stationary moment bound of Lemma 3 of Paper I, which does not rely on the series, needs every level of positive charge to have a charge-lowering jump; this holds for the Λ \Lambda Λ scheme with a type-(a) charge but not otherwise, so the statements above are statements about the pump series.)
Leading coefficients from the exact series. Throughout, an “exact” coefficient means one obtained from the exact finite-system formulation — a finite linear system whose solution is the Taylor coefficient itself, with no closure, fit, or radial truncation error — and evaluated numerically in double precision; it is neither a closed-form expression nor a value with a certified error bound, and its numerical reliability is assessed by the checks quoted below. In the fixed local basis the block-Jacobi matrix is affine in P P P , M ( P ) = M 0 + P M 1 \mathcal{M}(P)=\mathcal{M}_0+P\mathcal{M}_1 M ( P ) = M 0 + P M 1 (verified to $3\times10^{-14}$), and the finite-system series of § 3 B of Paper I applies: the coefficient vectors obey M 0 X ( r ) = − M 1 X ( r − 1 ) \mathcal{M}_0X^{(r)}=-\mathcal{M}_1X^{(r-1)} M 0 X ( r ) = − M 1 X ( r − 1 ) with vanishing trace component, X n ( r ) = 0 X_n^{(r)}=0 X n ( r ) = 0 for n > r n>r n > r , and observables are power series whose coefficients are ∑ m X 0 ( r ) [ m ] tr ( R m O ) \sum_m X_0^{(r)}[m]\operatorname{tr}(\mathcal{R}_mO) ∑ m X 0 ( r ) [ m ] tr ( R m O ) with P P P -independent traces. (In the damping basis the matrices are not affine in P P P , because the basis itself depends on the rates.) Table 2 gives the leading coefficients for $N=2,3 in the bad-cavity regime: all coefficients below the orders of Eq. (9) vanish to 4 × 10 − 14 4 \times10^{-14} 4 × 1 0 − 14 and all leading coefficients are nonzero, so the bounds are attained. Two features are worth noting. First, the leading coefficients of the Λ \Lambda Λ scheme and the ladder coincide for every observable in the table, and the ladder coefficients are independent of the pump-level decay rate (to eight digits for γ p / A = 0.3 \gamma_p/A = 0.3 γ p / A = 0.3 , 1 1 1 , 10 10 10 , with Λ \Lambda Λ as the limit γ p → ∞ \gamma_p \to\infty γ p → ∞ ); the two-level model and the V scheme share the leading photon-number and pair-coherence coefficients, but their leading higher connected correlations differ and depend on the reservoir decay of the V scheme (Γ 3 / P 2 = 0.675 \Gamma_3/P_2 = 0.675 Γ 3 / P 2 = 0.675 , 0.715 0.715 0.715 , 0.756 0.756 0.756 , 0.801 0.801 0.801 for γ r / A = 0.5 \gamma_r/A = 0.5 γ r / A = 0.5 , 1 1 1 , 2 2 2 , 10 10 10 , approaching the two-level 0.807 0.807 0.807 ; likewise K 3 / P 3 K_3/P_3 K 3 / P 3 ). We have not derived this pattern in general, but it is consistent with where the pump acts: in the Λ \Lambda Λ scheme and the ladder it acts only on a reservoir that carries no lasing coherence, whereas in the two-level and V schemes it empties the lower laser level and thereby acts on the coherences that the higher connected correlations are built from. Higher connected correlations thus retain information about the pump kinetics that the lower moments do not. Second, the leading pair-coherence coefficient of the Λ \Lambda Λ and ladder media is positive at these parameters, C 2 = + 0.106 ( P / A ) 2 C_2 = +0.106\,(P/A)^2 C 2 = + 0.106 ( P / A ) 2 , whereas that of the two-level and V schemes is negative, C 2 = − 1.14 ( P / A ) 2 C_2 = -1.14\,(P/A)^2 C 2 = − 1.14 ( P / A ) 2 at N = 2 N = 2 N = 2 : Paper I traced the negative two-level coherence to the trapping of population in the dark pair singlet, which only local pump and decay can enter or leave; a lower laser level emptied by recycling depletes the singlet, which requires one emitter in l l l . For the test posed in § 1 this is a qualitative failure: since ⟨ n ^ ⟩ \langle\hat n\rangle ⟨ n ^ ⟩ is linear in P P P at weak pump for every scheme, fitting the two-level pump rate to the photon number gives C 2 ≃ − 1.67 ⟨ n ^ ⟩ C_2 \simeq-1.67\langle\hat n\rangle C 2 ≃ − 1.67 ⟨ n ^ ⟩ for two two-level emitters, against C 2 ≃ + 0.079 ⟨ n ^ ⟩ C_2 \simeq+0.079\langle\hat n\rangle C 2 ≃ + 0.079 ⟨ n ^ ⟩ for two Λ \Lambda Λ or ladder emitters at these parameters — wrong in sign and in order.
Table 2. TABLE II. Leading weak-pump coefficients from the exact finite-system series in the bad-cavity regime (g = 1.1 A g=1.1A g = 1.1 A , γ = 0.3 A \gamma=0.3A γ = 0.3 A , γ r = γ p = A \gamma_r=\gamma_p=A γ r = γ p = A , ν = 0 \nu=0 ν = 0 ), with P P P in units of A A A : ⟨ n ^ ⟩ = n 1 P + … \langle\hat{n}\rangle=n_1P+\ldots ⟨ n ^ ⟩ = n 1 P + … , C 2 = c k P k + … C_2=c_kP^k+\ldots C 2 = c k P k + … , Γ 3 = γ k P k + … \Gamma_3=\gamma_kP^k+\ldots Γ 3 = γ k P k + … , K 3 = κ 3 P 3 + … K_3=\kappa_3P^3+\ldots K 3 = κ 3 P 3 + … . The largest coefficient below the leading order is 4 × 10 − 14 4\times10^{-14} 4 × 1 0 − 14 (two-level and V, N = 3 N=3 N = 3 ) and below 10 − 14 10^{-14} 1 0 − 14 otherwise; the affine residual of the fixed-basis matrix and the violation of X n ( r ) = 0 X_n^{(r)}=0 X n ( r ) = 0 for n > r n>r n > r are below 3 × 10 − 13 3\times10^{-13} 3 × 1 0 − 13 in every case.
Sign boundary. The filtration fixes the order of the pair coherence, not its sign, and the sign depends on the recycling rate. Figure 2 maps the leading coefficient c 2 c_2 c 2 of two Λ \Lambda Λ emitters, C 2 = c 2 P 2 + O ( P 3 ) C_2 = c_2P^2+O(P^3) C 2 = c 2 P 2 + O ( P 3 ) , as a function of γ r \gamma_r γ r , g g g , and γ \gamma γ (the ladder with γ p = γ r \gamma_p = \gamma_r γ p = γ r has the same c 2 c_2 c 2 ). At slow recycling the lower laser level is populated, the singlet traps population as for two-level emitters, and c 2 < 0 c_2 < 0 c 2 < 0 ; above a boundary γ r ∗ \gamma_r^* γ r ∗ the coherence is positive — a bright-triplet excess, i.e. constructive collective interference in the sense of Eq. (8) , not a statement about superradiant scaling of the output. The boundary is γ r ∗ ≈ γ \gamma_r^* \approx\gamma γ r ∗ ≈ γ when the laser-transition decay dominates (γ r ∗ / A = 10.0 \gamma_r^*/A = 10.0 γ r ∗ / A = 10.0 , 10.1 10.1 10.1 , 10.3 10.3 10.3 at γ = 10 A \gamma= 10A γ = 10 A for g / A = 0.5 g/A = 0.5 g / A = 0.5 , 1.1 1.1 1.1 , 2 2 2 ) and saturates at a coupling-dependent value when γ → 0 \gamma\to0 γ → 0 (γ r ∗ / A = 0.19 \gamma_r^*/A = 0.19 γ r ∗ / A = 0.19 , 0.54 0.54 0.54 , 1.04 1.04 1.04 ). At the good-cavity parameters of § 3 (g = 5 A g = 5A g = 5 A , γ = A \gamma= A γ = A ) the weak-pump boundary lies at γ r ∗ = 3.547 A \gamma_r^* = 3.547A γ r ∗ = 3.547 A , between the two ladders of Figure 3 : the fast one (γ r = 10 A \gamma_r = 10A γ r = 10 A ) has a positive and the slow one (γ r = 3 A \gamma_r = 3A γ r = 3 A ) a negative pair coherence over the whole pump range. The finite-pump sign change lies near this value and shifts with pump and emitter number, by at most 9 % 9\% 9% for the cases checked (N = 2 , 3 N = 2, 3 N = 2 , 3 up to P = 16 A P = 16A P = 16 A , where Figure 6 is computed; Appendix § 4.3 , item (ii)).
Figure 2. Leading pair-coherence coefficient c 2 c_2 c 2 of two Λ \Lambda Λ emitters, C 2 = c 2 P 2 + O ( P 3 ) C_2=c_2P^2+O(P^3) C 2 = c 2 P 2 + O ( P 3 ) , from the exact finite-system series (evaluated numerically) at Δ = 0 \Delta=0 Δ = 0 , ν = 0 \nu=0 ν = 0 : (a) versus recycling rate for several couplings and laser-transition decays (symmetric-log scale); (b) the sign boundary γ r ∗ \gamma_r^* γ r ∗ versus γ \gamma γ for three couplings; c 2 > 0 c_2>0 c 2 > 0 above the curves.
Figure 3. Pump-scheme dependence of two- and three-emitter lasing in the good-cavity regime g = 5 A g = 5A g = 5 A , γ = A \gamma= A γ = A , Δ = 0 \Delta= 0 Δ = 0 , ν = 0 \nu= 0 ν = 0 , versus pump rate: (a,e) mean photon number, (b,f) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) (insets: the region near unity), (c,g) pair coherence C 2 ( N ) C_2^{(N)} C 2 ( N ) , (d,h) width − 2 Re λ 1 -2\operatorname{Re}\lambda_1 − 2 Re λ 1 of the dominant spectral pole; top row N = 2 N = 2 N = 2 , bottom row N = 3 N = 3 N = 3 . Two-level model; V scheme with γ r = 20 A \gamma_r = 20A γ r = 20 A ; Λ \Lambda Λ scheme with γ r = 10 A \gamma_r = 10A γ r = 10 A ; ladder with γ p = γ r = 10 A \gamma_p = \gamma_r = 10A γ p = γ r = 10 A and with γ p = γ r = 3 A \gamma_p = \gamma_r = 3A γ p = γ r = 3 A . Thin grey lines mark g ( 2 ) ( 0 ) = 1 g^{(2)}(0) = 1 g ( 2 ) ( 0 ) = 1 and C 2 = 0 C_2 = 0 C 2 = 0 . All points are converged in n max n_{\max} n m a x (n max = 70 n_{\max} = 70 n m a x = 70 ; the largest-photon-number point of the scan, N = 3 N = 3 N = 3 , two-level, P / A = 32 P/A = 32 P / A = 32 , changes by 4 × 10 − 10 4 \times10^{-10} 4 × 1 0 − 10 between n max = 70 n_{\max} = 70 n m a x = 70 and 120 and agrees with a brute-force Fock-space solution to eight digits, Appendix § 4.3 ).
Figure 6. Dependence on the recycling rate at fixed pump P = 16 A P = 16A P = 16 A in the good-cavity regime, for the Λ \Lambda Λ scheme and the ladder (γ p = γ r \gamma_p = \gamma_r γ p = γ r ), N = 2 N = 2 N = 2 (solid) and N = 3 N = 3 N = 3 (dashed): (a) photon number, (b) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , (c) pair coherence, (d) dominant-pole width.
Strong pump: self-quenching versus saturation
Which schemes quench
The flux identity (9) gives a sufficient condition for strong-pump quenching.
Proposition 4 (Pump-induced quenching, ν = 0 \nu= 0 ν = 0 ). Let ν = 0 \nu= 0 ν = 0 and keep A > 0 A > 0 A > 0 , g g g , Δ \Delta Δ , and all pump-independent rates fixed. If the pump empties a laser level — P P P enters Γ u \Gamma_u Γ u or Γ l \Gamma_l Γ l , so that C u l → ∞ C_{ul} \to\infty C u l → ∞ as P → ∞ P \to\infty P → ∞ — then
⟨ n ^ ⟩ ≤ N 2 R u l A − N R u l , R u l ≤ g 2 A + 2 C u l , (10) \langle\hat n\rangle\le\frac{N^2\mathcal{R}_{ul}}{A-N\mathcal{R}_{ul}}, \qquad\mathcal{R}_{ul} \le\frac{g^2}{A+2C_{ul}}, \tag*{(10)} ⟨ n ^ ⟩ ≤ A − N R u l N 2 R u l , R u l ≤ A + 2 C u l g 2 , (10)
once N R u l < A N\mathcal{R}_{ul} < A N R u l < A , so that ⟨ n ^ ⟩ → 0 \langle\hat n\rangle\to0 ⟨ n ^ ⟩ → 0 as P → ∞ P \to\infty P → ∞ , in every Fock truncation and, under (U), for ρ s s \rho_{\mathrm{ss}} ρ ss .
Proof. In a truncation to Fock levels 0 , … , M − 1 0,\ldots,M-1 0 , … , M − 1 , with a compressed annihilation operator, [ n ^ , a † ] = a † [\hat n,a^\dagger]=a^\dagger [ n ^ , a † ] = a † and the cavity-loss terms of the equations for ⟨ n ^ ⟩ \langle\hat n\rangle ⟨ n ^ ⟩ and Z Z Z are unchanged, so the first line of Eq. (8) holds as it stands; the only change is [ a , a † ] = 1 − M Π M − 1 [a,a^\dagger]=1-M\Pi_{M-1} [ a , a † ] = 1 − M Π M − 1 , with Π M − 1 \Pi_{M-1} Π M − 1 the projector on the top Fock level, which adds − M ⟨ Π M − 1 S + S − ⟩ -M\langle\Pi_{M-1}S_+S_-\rangle − M ⟨ Π M − 1 S + S − ⟩ to the bracket in the second line. Since Π M − 1 \Pi_{M-1} Π M − 1 and S + S − S_+S_- S + S − are commuting positive operators, this term is nonpositive, and in every truncation as well as for the untruncated equation A ⟨ n ^ ⟩ ≤ R u l ⟨ S + S − + ⟨ n ^ ⟩ W u l ⟩ ≤ R u l [ N 2 + N ⟨ n ^ ⟩ ] A\langle\hat n\rangle\le\mathcal{R}_{ul}\langle S_+S_-+\langle\hat n\rangle W_{ul}\rangle\le\mathcal{R}_{ul}[N^2+N\langle\hat n\rangle] A ⟨ n ^ ⟩ ≤ R u l ⟨ S + S − + ⟨ n ^ ⟩ W u l ⟩ ≤ R u l [ N 2 + N ⟨ n ^ ⟩] , using ∥ S + S − ∥ ≤ N 2 \|S_+S_-\| \le N^2 ∥ S + S − ∥ ≤ N 2 and W u l ≤ N W_{ul} \le N W u l ≤ N commuting with n ^ ≥ 0 \hat n \ge0 n ^ ≥ 0 . The bound on R u l \mathcal{R}_{ul} R u l follows from Eq. (8) by dropping Δ 2 \Delta^2 Δ 2 . For ρ s s \rho_{\mathrm{ss}} ρ ss the identities hold because its moments are finite (Proposition 2(v)). □
For a thermal cavity, ν > 0 \nu> 0 ν > 0 , the analogous two-sided bounds on ⟨ n ^ ⟩ − ν \langle\hat n\rangle-\nu ⟨ n ^ ⟩ − ν give ⟨ n ^ ⟩ → ν \langle\hat n\rangle\to\nu ⟨ n ^ ⟩ → ν for the untruncated equation, but only for a stationary state with finite second moments, which hypothesis (U) does not provide; in a finite truncation the limit is instead the truncated thermal mean ν − M r M / ( 1 − r M ) \nu-Mr^M/(1-r^M) ν − M r M / ( 1 − r M ) , r = ν / ( 1 + ν ) r=\nu/(1+\nu) r = ν / ( 1 + ν ) , and the flux identity acquires boundary terms from the thermal pump. We do not pursue this case.
Pumping out of a laser level is a sufficient mechanism for quenching, not a necessary one. A pump can also starve the laser transition without broadening it, by diverting population into a shelving cycle: with levels a , b , u , l a,b,u,l a , b , u , l , pump a → b a\to b a → b at rate P P P , and fixed rates b → a b\to a b → a , b → u b\to u b → u , u → l u\to l u → l , l → a l\to a l → a all equal to A A A , the rate graph is irreducible and C u l = A C_{ul}=A C u l = A is pump independent, yet the single-emitter photon number at g = 1.1 A g=1.1A g = 1.1 A falls from 0.0675 0.0675 0.0675 at P = A P=A P = A to 2.5 × 10 − 4 2.5\times10^{-4} 2.5 × 1 0 − 4 at P = 1000 A P=1000A P = 1000 A , roughly as 1 / P 1/P 1/ P , because the feeding current a → u a\to u a → u vanishes as population accumulates in b b b . What saturation requires is that the pump does not reduce the injection current into the upper laser level; the Λ \Lambda Λ and ladder cycles studied here satisfy this.
For the Λ \Lambda Λ scheme and the ladder, C u l C_{ul} C u l is pump independent and the limit is a different model: as $P \to \infty the pumped level is emptied instantly, so the Λ \Lambda Λ scheme becomes the two-level model with pump rate γ r \gamma_r γ r (the recycling jump followed by instantaneous re-excitation) and the ladder becomes the V scheme with pump γ r \gamma_r γ r and reservoir decay γ p \gamma_p γ p . Numerically, at N = 2 N=2 N = 2 in the good-cavity regime, the Λ \Lambda Λ laser with γ r = 10 A \gamma_r=10A γ r = 10 A reaches the photon number 6.678 of its limiting two-level model to 10 − 3 10^{-3} 1 0 − 3 at P = 6400 A P=6400A P = 6400 A , and the ladder its V limit 4.342 likewise, while the two-level and V photon numbers fall as 1 / P 1/P 1/ P (Appendix § 4.3 , item (i)). The approach to the finite limit is monotonic for the cycles studied here; for other graphs neither monotonicity nor a nonzero limit is implied.
Exact comparison through the lasing crossover
Figure 3 follows N = 2 N=2 N = 2 and N = 3 N=3 N = 3 emitters through the lasing crossover for the four schemes at fixed laser-transition parameters and increasing pump rate. In the two-level model and the V scheme the photon number passes through a maximum (largest sampled values in Figure 3 : ⟨ n ^ ⟩ = 10.65 \langle\hat{n}\rangle=10.65 ⟨ n ^ ⟩ = 10.65 at P / A = 22 P/A=22 P / A = 22 for N = 2 N=2 N = 2 and 24.9 at P / A = 32 P/A=32 P / A = 32 for N = 3 N=3 N = 3 , 6.6 and 13.3 in the V scheme; maximizing over P P P gives 10.77 at P / A = 24.5 P/A=24.5 P / A = 24.5 and 25.36 at P / A = 36.7 P/A=36.7 P / A = 36.7 for the two-level model and 6.70 and 13.58 for the V scheme) and then collapses, to 2.3 and 12.5 (two-level) and 1.7 and 5.4 (V) at P / A = 64 P/A=64 P / A = 64 , while g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) rises back towards thermal values and the pair coherence, positive and growing through the lasing regime, falls back towards zero. This is the self-quenching of incoherently pumped one-atom lasers [1 ] , here with its collective signature, and Proposition 4 shows it is unavoidable for these two schemes at zero cavity temperature. In the Λ \Lambda Λ scheme and the ladder nothing quenches: the photon number, the inversion, and the linewidth saturate monotonically towards their limiting models, g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) passes through a shallow minimum near P / A ≈ 8 P/A\approx8 P / A ≈ 8 and stays within 1.0–1.1 for the Λ \Lambda Λ scheme and the fast ladder (minimum 0.996), and the pair coherence saturates, positive for γ r = 10 A \gamma_r=10A γ r = 10 A and negative over the whole range for the slow ladder (Sec. IV). At weak pump the Λ \Lambda Λ and ladder lasers are inverted from the start, because the lower laser level is emptied by recycling rather than filled by the pump: at P / A = 0.5 P/A = 0.5 P / A = 0.5 their photon numbers are roughly two to two and a half times those of the two-level model (0.57 and 0.54 against 0.28 at N = 2 N = 2 N = 2 ; 0.89 and 0.85 against 0.34 at N = 3 N = 3 N = 3 ) and their equal-time statistics are already near-Poissonian (g ( 2 ) ( 0 ) = 1.05 g^{(2)}(0) = 1.05 g ( 2 ) ( 0 ) = 1.05 , 1.01) where the two-level model is still thermal-like (1.6).
Comparisons at fixed bare pump rate are not comparisons at fixed excitation: the pump absorbed per emitter is J = P p o r i g i n J = Pp_{\rm origin} J = P p origin , with p o r i g i n p_{\rm origin} p origin the population of the level the pump empties, and the schemes differ in p o r i g i n p_{\rm origin} p origin . Figure 4 replots the same data against J J J , showing the conversion fraction η = A ⟨ n ^ ⟩ / ( N J ) \eta= A\langle\hat{n}\rangle/(NJ) η = A ⟨ n ^ ⟩ / ( N J ) , the share of the absorbed excitation current that leaves through the cavity — an excitation-current conversion fraction, not a pump-power or wall-plug efficiency, since neither the pump photon energy nor the losses of the pump source enter it: by the first line of Eq. (8) , A ⟨ n ^ ⟩ = N J − γ ⟨ N u ⟩ A\langle\hat{n}\rangle= NJ - \gamma\langle N_u\rangle A ⟨ n ^ ⟩ = N J − γ ⟨ N u ⟩ in stationarity, and 1 − η 1 - \eta 1 − η is the share lost by spontaneous decay of the laser transition. Matching the absorbed current reduces some of the differences seen at fixed pump rate, but appreciable ones remain because the spontaneous loss depends on the level scheme. At J = A J = A J = A the photon numbers are 1.112, 1.142, 1.481, 1.478 (two-level, V, Λ \Lambda Λ , fast ladder) for N = 2 N = 2 N = 2 and 1.604, 1.655, 2.321, 2.319 for N = 3 N = 3 N = 3 , i.e. η = 0.56 \eta= 0.56 η = 0.56 , 0.57, 0.74, 0.74 and 0.53, 0.55, 0.77, 0.77: the
Figure 4. The data of Fig. 3 against the absorbed pump current per emitter J = P o r i g i n J = P_{\mathrm{origin}} J = P origin : (a) conversion fraction η = A ⟨ n ^ ⟩ / ( N J ) \eta= A\langle\hat{n}\rangle/(NJ) η = A ⟨ n ^ ⟩ / ( N J ) , (b) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , (c) pair coherence; thick lines N = 2 N = 2 N = 2 , thin lines N = 3 N = 3 N = 3 . The two-level and V curves turn around because J J J decreases again at strong pump.
for N = 2 N = 2 N = 2 and 1.604, 1.655, 2.321, 2.319 for N = 3 N = 3 N = 3 , i.e. η = 0.56 \eta= 0.56 η = 0.56 , 0.57, 0.74, 0.74 and 0.53, 0.55, 0.77, 0.77: the conversion fraction of the Λ \Lambda Λ and ladder lasers exceeds that of the two-level laser by 33% and 45% at N = 2 N = 2 N = 2 and 3, because their lower laser level is kept empty by recycling and the cavity competes with γ r \gamma_r γ r from an inverted transition. Quenching appears in Figure 4 as the absorbed current of the two-level and V schemes turning back once the lower laser level is emptied: it reaches a maximum of J / A = 6.1 J/A = 6.1 J / A = 6.1 and 9.2 for the two-level laser and 4.0 and 5.1 for the V laser (N = 2 , 3 N = 2, 3 N = 2 , 3 ), while the Λ \Lambda Λ and ladder currents keep growing to the recycling bottleneck. Figure 3 is therefore a parameter-matched illustration, not an efficiency ranking.
Linewidth and recycling
The dominant-pole width (Figure 3 d,Figure 3 h; the selection of the dominant pole is insensitive to the residue threshold, Appendix D) is, over the lasing range, well described by its inverse proportionality to the photon number, so that at equal P P P the scheme with the larger ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ has the narrower line: at P / A = 16 P/A = 16 P / A = 16 the widths are 0.108 A 0.108A 0.108 A , 0.138 A 0.138A 0.138 A , 0.241 A 0.241A 0.241 A , 0.292 A 0.292A 0.292 A for the two-level, V, Λ \Lambda Λ , and ladder (γ p = γ r = 10 A \gamma_p = \gamma_r = 10A γ p = γ r = 10 A ) schemes at N = 2 N = 2 N = 2 , and 0.077 A 0.077A 0.077 A , 0.091 A 0.091A 0.091 A , 0.182 A 0.182A 0.182 A , 0.212 A 0.212A 0.212 A at N = 3 N = 3 N = 3 . The schemes that do not quench do not have narrower lines at fixed pump in this regime; their advantage is that the line does not rebroaden at strong pump, where the two-level and V linewidths pass through a minimum (0.078 A 0.078A 0.078 A and 0.119 A 0.119A 0.119 A at N = 2 N = 2 N = 2 , 0.027 A 0.027A 0.027 A and 0.053 A 0.053A 0.053 A at N = 3 N = 3 N = 3 ) and then grow again through the quenching transition, by a factor of three to four at N = 2 N = 2 N = 2 and two to three at N = 3 N = 3 N = 3 by P / A = 64 P/A = 64 P / A = 64 . Figure 5 shows the spectral functions at P / A = 16 P/A = 16 P / A = 16 and 64 64 64 : at P / A = 16 P/A = 16 P / A = 16 the two-level line is the narrowest (full width at half maximum 0.100 A 0.100A 0.100 A , against 0.133 A 0.133A 0.133 A , 0.233 A 0.233A 0.233 A , 0.283 A 0.283A 0.283 A ), while at P / A = 64 P/A = 64 P / A = 64 the order is reversed, the quenched two-level and V lines having broadened to 0.29 A 0.29A 0.29 A and 0.37 A 0.37A 0.37 A while the Λ \Lambda Λ and ladder lines, 0.22 A 0.22A 0.22 A and 0.27 A 0.27A 0.27 A , are essentially unchanged. The FWHM agrees with the dominant-pole width to 3–7% at P / A = 16 P/A = 16 P / A = 16 and 0.5–3% at P / A = 64 P/A = 64 P / A = 64 .
Figure 5. Normalized emission spectra S ( ω ′ ) S(\omega') S ( ω ′ ) of the N = 2 N = 2 N = 2 lasers of Fig. 3 at (a) P / A = 16 P/A = 16 P / A = 16 and (b) P / A = 64 P/A = 64 P / A = 64 , versus detuning from the cavity frequency, from the pole decomposition of the K = − 1 K = -1 K = − 1 continued fraction (n max = 60 n_{\max} = 60 n m a x = 60 ).
For the Λ \Lambda Λ scheme and the ladder the recycling rate plays the role that P P P plays for the V scheme in C u l C_{\mathrm{ul}} C ul : it must be fast enough to empty the lower laser level but not so fast as to dephase the coherence. Figure 6 scans it at fixed P = 16 A P = 16A P = 16 A . Photon number and linewidth are both optimized at recycling rates well above the laser-transition decay and comparable to or above the pump, γ r / A ≈ 16 \gamma_r/A \approx16 γ r / A ≈ 16 –32, but not at the same value: the N = 2 N = 2 N = 2 Λ \Lambda Λ laser reaches ⟨ n ^ ⟩ = 6.05 \langle\hat{n}\rangle= 6.05 ⟨ n ^ ⟩ = 6.05 at γ r / A = 16 \gamma_r/A = 16 γ r / A = 16 and its narrowest line, 0.129 A 0.129A 0.129 A , at γ r / A = 22 \gamma_r/A = 22 γ r / A = 22 ; the N = 3 N = 3 N = 3 Λ \Lambda Λ laser 12.1 at γ r / A = 22 \gamma_r/A = 22 γ r / A = 22 and 0.060 A 0.060A 0.060 A at γ r / A = 32 \gamma_r/A = 32 γ r / A = 32 ; the ladder (γ p = γ r \gamma_p = \gamma_r γ p = γ r ) 4.55 4.55 4.55 and 0.167 A 0.167A 0.167 A at γ r / A = 22 \gamma_r/A = 22 γ r / A = 22 for N = 2 N = 2 N = 2 , and 9.24 9.24 9.24 and 0.076 A 0.076A 0.076 A at γ r / A = 32 \gamma_r/A = 32 γ r / A = 32 for N = 3 N = 3 N = 3 . On either side the laser degrades for opposite reasons: at slow recycling the lower laser level fills, the inversion is lost, the statistics become bunched (g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) up to 1.4 at N = 3 N = 3 N = 3 ), and the pair coherence turns negative below γ r / A = 3.5 \gamma_r/A = 3.5 γ r / A = 3.5 (N = 2 N = 2 N = 2 ) and 3.25 (N = 3 N = 3 N = 3 ), near the weak-pump boundary of Figure 2 (Sec. IV); at fast recycling C u l = ( γ + γ r ) / 2 C_{\mathrm{ul}} = (\gamma+ \gamma_r)/2 C ul = ( γ + γ r ) /2 dephases the coherence, the photon number collapses, and the field returns to thermal statistics, g ( 2 ) ( 0 ) → 2 g^{(2)}(0) \to2 g ( 2 ) ( 0 ) → 2 . The exact solution thus allows the recycling rate of a given few-emitter device to be optimized for photon number or for linewidth.
Photon statistics and the pump cycle
A value g ( 2 ) ( 0 ) < 1 g^{(2)}(0) < 1 g ( 2 ) ( 0 ) < 1 establishes a sub-Poissonian intracavity photon distribution, F n = Var ( n ) / ⟨ n ⟩ = 1 + ⟨ n ^ ⟩ / [ g ( 2 ) ( 0 ) − 1 ] < 1 F_n = \operatorname{Var}(n)/\langle n\rangle= 1 + \langle\hat{n}\rangle/[g^{(2)}(0)-1] < 1 F n = Var ( n ) / ⟨ n ⟩ = 1 + ⟨ n ^ ⟩ / [ g ( 2 ) ( 0 ) − 1 ] < 1 . It does not by itself establish reduced noise in the detected output, which depends on the time integral of g ( 2 ) ( τ ) − 1 g^{(2)}(\tau)-1 g ( 2 ) ( τ ) − 1 : for ideal detection of the cavity-loss channel, the Fano factor of the photon count in a long interval is
F c o u n t = 1 + 2 A ⟨ n ^ ⟩ ∫ 0 ∞ [ g ( 2 ) ( τ ) − 1 ] d τ . (11) F_{\mathrm{count}} = 1 + 2A\langle\hat{n}\rangle\int_{0}^{\infty} [g^{(2)}(\tau)-1]\,d\tau. \tag*{(11)} F count = 1 + 2 A ⟨ n ^ ⟩ ∫ 0 ∞ [ g ( 2 ) ( τ ) − 1 ] d τ . (11)
With a finite overall detection efficiency η \eta η (collection and detector), the counts are a binomial thinning of the emitted photons and the measured Fano factor is 1 + η ( F c o u n t − 1 ) 1+\eta(F_{\mathrm{count}}-1) 1 + η ( F count − 1 ) : the suppression is reduced in proportion to η \eta η but keeps its sign. Both are available exactly. The intensity correlation ⟨ a † ( 0 ) a † ( τ ) a ( τ ) a ( 0 ) ⟩ = tr [ n ^ e L τ ( a ρ s s a † ) ] \langle a^\dagger(0)a^\dagger(\tau)a(\tau)a(0)\rangle= \operatorname{tr}[\hat{n}e^{\mathcal{L}\tau}(a\rho_{\mathrm{ss}}a^\dagger)] ⟨ a † ( 0 ) a † ( τ ) a ( τ ) a ( 0 )⟩ = tr [ n ^ e L τ ( a ρ ss a † )] evolves the permutation-invariant, charge-zero operator a ρ s s a † a\rho_{\mathrm{ss}}a^\dagger a ρ ss a † under the K = 0 K = 0 K = 0 block-Jacobi operator; its coefficients follow from those of ρ s s \rho_{\mathrm{ss}} ρ ss by two field identities, the functional tr ( n ^ ⋅ ) \operatorname{tr}(\hat{n}\cdot) tr ( n ^ ⋅ ) picks the n = 1 n = 1 n = 1 , N e 0 Ne_0 N e 0 component, and the pole decomposition of the truncated K = 0 K = 0 K = 0 operator gives g ( 2 ) ( τ ) = ∑ j w j e λ j τ / ⟨ n ^ ⟩ 2 g^{(2)}(\tau) = \sum_j w_j e^{\lambda_j\tau}/\langle\hat{n}\rangle^2 g ( 2 ) ( τ ) = ∑ j w j e λ j τ / ⟨ n ^ ⟩ 2 , with the λ = 0 \lambda= 0 λ = 0 term equal to ⟨ n ^ ⟩ 2 \langle\hat{n}\rangle^2 ⟨ n ^ ⟩ 2 and F c o u n t = 1 + ( 2 A / ⟨ n ^ ⟩ ) ∑ λ j ≠ 0 w j / ( − λ j ) F_{\mathrm{count}} = 1 + (2A/\langle\hat{n}\rangle)\sum_{\lambda_j\ne0} w_j/(-\lambda_j) F count = 1 + ( 2 A / ⟨ n ^ ⟩) ∑ λ j = 0 w j / ( − λ j ) . Because the residues of a non-normal operator carry rounding errors, we evaluate the integral in Eq. (11) instead from the Poisson equation M Z = − ( Y − ⟨ n ^ ⟩ X s s ) \mathcal{M}Z = -(Y-\langle\hat{n}\rangle X_{\mathrm{ss}}) M Z = − ( Y − ⟨ n ^ ⟩ X ss ) , tr Z = 0 \operatorname{tr}Z = 0 tr Z = 0 , for the truncated K = 0 K = 0 K = 0 block-Jacobi matrix M \mathcal{M} M , the coefficient vector Y Y Y of a ρ s s a † a\rho_{\mathrm{ss}}a^\dagger a ρ ss a † , and the stationary vector X s s X_{\mathrm{ss}} X ss ; this is independent of the radial cutoff to nine digits, whereas the pole sum scatters at the 10 − 6 10^{-6} 1 0 − 6 level. An independent calculation in the full Fock⊗ \otimes ⊗ emitter space (no symmetry reduction, no damping basis, Poisson equation for the Liouvillian) reproduces F c o u n t F_{\mathrm{count}} F count to 10 − 12 10^{-12} 1 0 − 12 or better for one, two, and three emitters (Appendix § 4.3 ).
Figure 3b,f already shows that the ladder with slow steps, γ p = γ r = 3 A \gamma_p = \gamma_r = 3A γ p = γ r = 3 A , has g ( 2 ) ( 0 ) < 1 g^{(2)}(0) < 1 g ( 2 ) ( 0 ) < 1 over the weak-to-intermediate pump range (P ≲ 4 A P \lesssim4A P ≲ 4 A at N = 2 N = 2 N = 2 , P ≲ 0.7 A P \lesssim0.7A P ≲ 0.7 A at N = 3 N = 3 N = 3 , where ⟨ n ^ ⟩ ≲ 1 \langle\hat{n}\rangle\lesssim1 ⟨ n ^ ⟩ ≲ 1 ), 0.849 at P = 0.3 A P = 0.3A P = 0.3 A for N = 2 N = 2 N = 2 and 0.955 for N = 3 N = 3 N = 3 , and that the two-level, V, and Λ \Lambda Λ schemes of that figure stay above one throughout. Such a comparison at fixed pump rate does not isolate the mechanism, because the schemes have different photon numbers and different absorbed currents. Table III therefore makes the test of § 1 explicit. Held fixed in every row are the cavity (A A A , Δ = 0 \Delta= 0 Δ = 0 , ν = 0 \nu= 0 ν = 0 ), the coupling g = 5 A g = 5A g = 5 A , the laser-transition decay γ = A \gamma= A γ = A , and N N N ; the only fitted parameter is the pump rate P P P , chosen so that every scheme gives the same ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ ; the observables compared after the fit are g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , F n F_n F n , and F c o u n t F_{\mathrm{count}} F count . It compares the ladder with a fixed recycling rate γ r = 3 A \gamma_r = 3A γ r = 3 A and a pump-level decay γ p \gamma_p γ p that is varied from slow to fast — γ p → ∞ \gamma_p \to\infty γ p → ∞ is the Λ \Lambda Λ scheme with the ground level as reservoir — against the V scheme with the same γ r \gamma_r γ r , the V scheme with fast reservoir decay, and the two-level model, for N = 1 , 2 , 3 N = 1,2,3 N = 1 , 2 , 3 and ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle= 0.3 ⟨ n ^ ⟩ = 0.3 and 1.
Table III. Photon statistics at matched photon number in the good-cavity regime (g = 5 A g=5A g = 5 A , γ = A \gamma=A γ = A ): pump rate P P P at which ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ equals the target — the smallest such P P P , i.e. the rising branch below the photon-number maximum of the two-level and V schemes, found by bisection after a coarse scan — g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , intracavity Fano factor F n F_n F n , and long-time output-counting Fano factor F c o u n t F_{\mathrm{count}} F count , Eq. (11) . Ladder: γ r = 3 A \gamma_r=3A γ r = 3 A and the pump-level decay γ p \gamma_p γ p indicated; Λ \Lambda Λ : γ r = 3 A \gamma_r=3A γ r = 3 A (the γ p → ∞ \gamma_p\to\infty γ p → ∞ limit of this ladder); V: γ r = 3 A \gamma_r=3A γ r = 3 A or 20 A 20A 20 A . At ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 the intracavity Fano factor F n = 1 + ⟨ n ^ ⟩ [ g ( 2 ) ( 0 ) − 1 ] F_n=1+\langle\hat{n}\rangle[g^{(2)}(0)-1] F n = 1 + ⟨ n ^ ⟩ [ g ( 2 ) ( 0 ) − 1 ] equals g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) and is not listed. “—”: the target photon number is not reached at any pump.
The table separates the two effects. At matched ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle= 0.3 ⟨ n ^ ⟩ = 0.3 , every statistic decreases monotonically as the pump step is slowed (at $\langle\hat{n}\rangle = 1 the same holds except for the last step γ p = 3 A → A \gamma_p = 3A \to A γ p = 3 A → A at N = 3 N = 3 N = 3 , where the three statistics rise by 0.002–0.007): for N = 2 N = 2 N = 2 at ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle= 0.3 ⟨ n ^ ⟩ = 0.3 , F c o u n t F_{\mathrm{count}} F count falls from 0.855 ( Λ \Lambda Λ , no pump-level bottleneck) through 0.852, 0.829, 0.775 to 0.663 as γ p \gamma_p γ p goes from $100A$ to A A A , and for N = 3 N = 3 N = 3 from 0.905 to 0.754; at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle= 1 ⟨ n ^ ⟩ = 1 , where g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) is within a few percent of one, the counting Fano factor is still 0.74–0.82 for N = 2 N = 2 N = 2 and 0.75–0.86 for $N = 3, so the output counting noise is reduced well into the regime where the equal-time statistics are near-Poissonian. The reference schemes order the same way: the V scheme with a slow reservoir decay (γ r = 3 A \gamma_r = 3A γ r = 3 A ), a two-step cycle, is quieter than the V scheme with γ r = 20 A \gamma_r = 20A γ r = 20 A , which is in turn quieter than the two-level model, whose excitation of the upper laser level is a single Poissonian jump — but for N ≥ 2 N \ge2 N ≥ 2 the two-level model and the fast V scheme stay above F c o u n t = 1 F_{\mathrm{count}} = 1 F count = 1 in every row, the slow V scheme dips below it only at N = 2 N = 2 N = 2 , ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle= 1 ⟨ n ^ ⟩ = 1 (0.97), whereas every ladder and the Λ \Lambda Λ scheme is below one in every row. (A single two-level emitter has F c o u n t = 0.93 F_{\mathrm{count}} = 0.93 F count = 0.93 at ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle= 0.3 ⟨ n ^ ⟩ = 0.3 although its $g^{(2)}(0) = 1.03: the counting statistics integrates g ( 2 ) ( τ ) − 1 g^{(2)}(\tau)-1 g ( 2 ) ( τ ) − 1 over the emitter’s recovery time.) The picture that emerges is the one of dynamic pump-noise suppression [10 , 11 , 12 ] , the multilevel counterpart of regular pumping [8 , 9 ] : rate-limited steps in the excitation cycle make the arrival of excitations in the upper laser level more regular than a Poisson process, and a long-lived cavity field inherits the regularity. Here it is obtained exactly for a few emitters from the full level kinetics, rather than from an effective pump operator; the treatment of Ref. [12 ] , which derives a non-Markovian, photon-number-dependent pump operator for a single multilevel atom, is the one-atom counterpart. Figure 7 shows the content of Table III together with a reference that isolates what the common cavity does.
Figure 7. The matched-photon-number comparison of Table III (g = 5 A g = 5A g = 5 A , γ = A \gamma= A γ = A ): (a,b) long-time output-counting Fano factor and (c,d) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) at ⟨ n ^ ⟩ = 0.3 \langle\hat{n} \rangle= 0.3 ⟨ n ^ ⟩ = 0.3 (left) and 1 (right), for the ladder with γ r = 3 A \gamma_r = 3A γ r = 3 A and pump-level decay γ p \gamma_p γ p (connected points; γ p → ∞ \gamma_p \to\infty γ p → ∞ is the Λ \Lambda Λ scheme) and for the reference schemes (V with γ r = 3 A \gamma_r = 3A γ r = 3 A , V with γ r = 20 A \gamma_r = 20A γ r = 20 A , two-level), for N = 1 , 2 , 3 N = 1, 2, 3 N = 1 , 2 , 3 emitters. Open symbols (dashed): N N N independent single-emitter lasers of the same scheme, each in its own cavity, at the same total photon number — the counting Fano factor of their combined output is that of one of them, and g t o t ( 2 ) ( 0 ) = [ g 1 ( 2 ) ( 0 ) + N − 1 ] / N g_{\mathrm{tot}}^{(2)}(0) = [g_1^{(2)}(0) + N - 1]/N g tot ( 2 ) ( 0 ) = [ g 1 ( 2 ) ( 0 ) + N − 1 ] / N . Missing points: the target photon number is not reached. Thin lines mark unity.
a. What the common cavity changes. Noise suppression by a rate-limited pump cycle is a single-emitter prop erty, and the counting statistics of independent sources superpose: for N N N independent stationary sources whose output photons are counted together — summed count streams, as from N N N cavities on one detector or N N N detectors whose counts are added, not coherent recombination of the fields into one detected mode — the long-time Fano factor of the combined count is the rate-weighted mean of the individual ones, and the equal-time correlation of the summed intensities is g t o t ( 2 ) ( 0 ) = [ g 1 ( 2 ) ( 0 ) + N − 1 ] / N g_{\mathrm{tot}}^{(2)}(0)=[g_1^{(2)}(0)+N-1]/N g tot ( 2 ) ( 0 ) = [ g 1 ( 2 ) ( 0 ) + N − 1 ] / N . The natural reference for N N N emitters in one cavity is therefore N N N copies of the single-emitter laser of the same scheme, each in its own cavity with the same A A A and g g g , driven so that the total photon number — and hence the mean output rate — is the same, each at the smallest pump rate giving ⟨ n ^ ⟩ 1 = ⟨ n ^ ⟩ / N \langle\hat{n}\rangle_1=\langle\hat{n}\rangle/N ⟨ n ^ ⟩ 1 = ⟨ n ^ ⟩ / N , the same rising-branch rule as in Table III (open symbols in Fig. 7; data in Ref. [27 ] ). In every row the common cavity makes the output noisier than the independent reference. For the Λ \Lambda Λ and ladder lasers the difference is small at ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle=0.3 ⟨ n ^ ⟩ = 0.3 (F c o u n t = 0.663 F_{\mathrm{count}}=0.663 F count = 0.663 against 0.640 for the slowest ladder at N = 2 N=2 N = 2 ; 0.855 against 0.839 for Λ \Lambda Λ ) and appreciable at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 (0.740 against 0.553 for the γ p = 3 A \gamma_p=3A γ p = 3 A ladder at N = 2 N=2 N = 2 ; 0.820 against 0.658 for Λ \Lambda Λ ; 0.858 against 0.709 for Λ \Lambda Λ at N = 3 N=3 N = 3 ), where the independent single-emitter lasers run at ⟨ n ^ ⟩ 1 = 1 / N \langle\hat{n}\rangle_1=1/N ⟨ n ^ ⟩ 1 = 1/ N each and keep more of their antibunching. For the two-level and V lasers the common cavity is what pushes the output above the shot-noise level: their independent references are below one in every row (0.93–0.96 for the two-level laser) while the common-cavity values exceed one in all but one row (0.97–1.31). The equal-time statistics tell the same story more strongly: the near-Poissonian g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) of the Λ \Lambda Λ and ladder lasers at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 (0.97–1.04) is a common-cavity effect, the independent reference being 0.85–0.91. Thus the sub-Poissonian counting statistics of the multilevel lasers is not evidence of a collective enhancement — the common cavity erodes a single-emitter effect, by about 0.02 at ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle=0.3 ⟨ n ^ ⟩ = 0.3 and by 0.15–0.20 at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 for the Λ \Lambda Λ and ladder cycles, so that between about half and 94% of the single-emitter suppression 1 − F c o u n t 1-F_{\mathrm{count}} 1 − F count survives (49% for Λ \Lambda Λ at N = 3 N=3 N = 3 , ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 ; 94% for the slowest ladder at
N = 2 N = 2 N = 2 , ⟨ n ^ ⟩ = 0.3 \langle\hat{n} \rangle= 0.3 ⟨ n ^ ⟩ = 0.3 ) — but the counting statistics reveals substantial suppression that is much less apparent in the equal-time statistics. The comparison is an operational one between two geometries at equal output: besides the inter-emitter correlations, the occupation of each cavity and with it the stimulated dynamics differ between the two arrangements, and the difference should not be read as a measure of the correlations alone. (Comparing at the same pump rate instead of the same total photon number changes the independent reference values by at most 0.02 for the Λ \Lambda Λ and ladder schemes; for the two-level and V schemes the independent lasers then have 1.1–1.9 times the photon number.)
Where the suppression sits in time
Integrated statistics hide the time scale on which the regularity acts. Figure 8 shows it for the headline case of Table III, two ladder emitters with γ p = 3 A \gamma_p = 3A γ p = 3 A at ⟨ n ^ ⟩ = 1 \langle\hat{n} \rangle= 1 ⟨ n ^ ⟩ = 1 , together with its Λ \Lambda Λ limit, the two-level model, and the matched independent references, through g ( 2 ) ( τ ) − 1 g^{(2)}(\tau)-1 g ( 2 ) ( τ ) − 1 and the cumulative integral
Figure 8. Time-resolved origin of the counting-noise suppression, for N = 2 N=2 N = 2 emitters at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 and the pump rates of Table III (g = 5 A g=5A g = 5 A , γ = A \gamma=A γ = A , γ r = 3 A \gamma_r=3A γ r = 3 A ): the ladder with γ p = 3 A \gamma_p=3A γ p = 3 A (F c o u n t = 0.740 F_{\mathrm{count}}=0.740 F count = 0.740 ), its Λ \Lambda Λ limit, and the two-level model. (a) g ( 2 ) ( τ ) − 1 g^{(2)}(\tau)-1 g ( 2 ) ( τ ) − 1 ; (b) cumulative counting integral I ( T ) I(T) I ( T ) , Eq. (12) , whose limit is F c o u n t − 1 F_{\mathrm{count}}-1 F count − 1 (markers at the right edge: the Poisson-equation values). Solid: common cavity; dashed: the matched independent-source reference (two single-emitter lasers at ⟨ n ^ ⟩ 1 = 1 / 2 \langle\hat{n}\rangle_1=1/2 ⟨ n ^ ⟩ 1 = 1/2 , summed count streams), for which g t o t ( 2 ) ( τ ) − 1 = [ g 1 ( 2 ) ( τ ) − 1 ] / 2 g_{\mathrm{tot}}^{(2)}(\tau)-1=[g_1^{(2)}(\tau)-1]/2 g tot ( 2 ) ( τ ) − 1 = [ g 1 ( 2 ) ( τ ) − 1 ] /2 and I t o t = I 1 I_{\mathrm{tot}}=I_1 I tot = I 1 . Computed by propagation with the truncated K = 0 K=0 K = 0 block-Jacobi matrix (n max = 30 n_{\max}=30 n m a x = 30 ; curves change by less than 4 × 10 − 14 4\times10^{-14} 4 × 1 0 − 14 at n max = 40 n_{\max}=40 n m a x = 40 ), not from residues.
I ( T ) = 2 A ⟨ n ^ ⟩ ∫ 0 T [ g ( 2 ) ( τ ) − 1 ] d τ , I ( ∞ ) = F c o u n t − 1. (12) I(T) = 2A\langle\hat{n} \rangle\int_{0}^{T} [g^{(2)}(\tau)-1]\,d\tau,\qquad I(\infty) = F_{\mathrm{count}}-1. \tag*{(12)} I ( T ) = 2 A ⟨ n ^ ⟩ ∫ 0 T [ g ( 2 ) ( τ ) − 1 ] d τ , I ( ∞ ) = F count − 1. (12)
Both follow from the K = 0 K = 0 K = 0 recurrence without residues: g ( 2 ) ( τ ) − 1 g^{(2)}(\tau)-1 g ( 2 ) ( τ ) − 1 is the trace functional of e M τ ( Y − ⟨ n ^ ⟩ X s s ) e^{M\tau}(Y-\langle\hat{n}\rangle X_{\mathrm{ss}}) e M τ ( Y − ⟨ n ^ ⟩ X ss ) , and I ( T ) = ( 2 A / ⟨ n ^ ⟩ ) tr [ n ^ ( 1 − e M T ) Z ] I(T) = (2A/\langle\hat{n}\rangle)\operatorname{tr}[\hat{n}(1-e^{M T})Z] I ( T ) = ( 2 A / ⟨ n ^ ⟩) tr [ n ^ ( 1 − e M T ) Z ] with the Poisson solution Z Z Z above. The anticorrelation is not at zero delay. For the ladder, g ( 2 ) ( τ ) g^{(2)}(\tau) g ( 2 ) ( τ ) falls from 0.974 at τ = 0 \tau= 0 τ = 0 to a minimum of 0.879 at A τ ≈ 0.43 A\tau\approx0.43 A τ ≈ 0.43 ; the Λ \Lambda Λ laser is slightly bunched at zero delay (1.010) and has its minimum, 0.911, at A τ ≈ 0.45 A\tau\approx0.45 A τ ≈ 0.45 . The negative area accumulates over two to three cavity lifetimes — 90% of I ( ∞ ) I(\infty) I ( ∞ ) is reached by A T ≈ 2.3 AT \approx2.3 A T ≈ 2.3 for both lasers — and I ( T ) I(T) I ( T ) settles at − 0.260 -0.260 − 0.260 and − 0.181 -0.181 − 0.181 , the values F c o u n t − 1 F_{\mathrm{count}}-1 F count − 1 of Table III. The minimum lies at a delay of the order of, though shorter than, the mean time an emitter needs to return from the lower to the upper laser level ( 1 / γ r + 1 / P + 1 / γ p ≈ 1 / A (1/\gamma_r + 1/P + 1/\gamma_p \approx1/A ( 1/ γ r + 1/ P + 1/ γ p ≈ 1/ A for the ladder at this pump, 1 / γ r + 1 / P ≈ 1 / A 1/\gamma_r + 1/P \approx1/A 1/ γ r + 1/ P ≈ 1/ A for Λ \Lambda Λ ); since this time and the cavity lifetime coincide here, Figure 8 does not separate the two. The independent references have deeper minima. that already begin at τ = 0 \tau=0 τ = 0 (g t o t ( 2 ) ( 0 ) = 0.852 g_{\mathrm{tot}}^{(2)}(0)=0.852 g tot ( 2 ) ( 0 ) = 0.852 and 0.894 0.894 0.894 ), with the minimum at a similar delay (A τ ≈ 0.34 A\tau\approx0.34 A τ ≈ 0.34 –0.36 0.36 0.36 ): the common cavity makes the dip shallower, removes most of it near zero delay, and moves its minimum only slightly to longer delay; their I ( T ) I(T) I ( T ) settles at the single-emitter values − 0.447 -0.447 − 0.447 and − 0.342 -0.342 − 0.342 . The two-level laser, by contrast, has g ( 2 ) ( τ ) > 1 g^{(2)}(\tau)>1 g ( 2 ) ( τ ) > 1 at all delays and accumulates a positive I ( ∞ ) = + 0.248 I(\infty)=+0.248 I ( ∞ ) = + 0.248 , although its own independent reference has a small negative area (− 0.068 -0.068 − 0.068 ). This is why equal-time statistics understate the effect of the pump cycle: g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) samples the correlation at the one delay at which the common-cavity dip has not yet formed.
c. What the fitted two-level pump gets wrong. With the parameters held fixed as stated above and only P P P fitted, the two-level model reproduces ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ by construction and nothing else in Table III : for N = 2 N=2 N = 2 at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle=1 ⟨ n ^ ⟩ = 1 it gives g ( 2 ) ( 0 ) = 1.28 g^{(2)}(0)=1.28 g ( 2 ) ( 0 ) = 1.28 and F c o u n t = 1.25 F_{\mathrm{count}}=1.25 F count = 1.25 against 0.97 0.97 0.97 –1.01 1.01 1.01 and 0.74 0.74 0.74 –0.82 0.82 0.82 for the ladder family, and for N = 2 , 3 N=2,3 N = 2 , 3 it places the output above the shot-noise level in every row where all ladders and the Λ \Lambda Λ scheme are below it.
Dark sectors
For two-level emitters, Paper I proved (its Proposition 8) that the collective coupling annihilates the minimal collective-spin sector if and only if N N N is even. For d ≥ 3 d\geq3 d ≥ 3 , S ± S_{\pm} S ± act as the collective pseudo-spin ladder of the emitters in laser levels and as zero on the others. Let n a n_a n a be the number of emitters in auxiliary levels and J J J the total pseudo-spin of the remaining M = N − n a M=N-n_a M = N − n a ; the projectors Π n a , J \Pi_{n_a,J} Π n a , J commute with permutations and sum to the identity, and for pooled auxiliary levels the sector has dimension ( N n a ) ( d − 2 ) n a ( 2 J + 1 ) m ( M , J ) \binom{N}{n_a}(d-2)^{n_a}(2J+1)m(M,J) ( n a N ) ( d − 2 ) n a ( 2 J + 1 ) m ( M , J ) , with m ( M , J ) = ( ( M M / 2 − J ) − ( M M / 2 − J − 1 ) ) m(M,J)=\left(\binom{M}{M/2-J}-\binom{M}{M/2-J-1}\right) m ( M , J ) = ( ( M /2 − J M ) − ( M /2 − J − 1 M ) ) . A sector is annihilated by both S ± S_{\pm} S ± exactly when J = 0 J=0 J = 0 , so for every N ≥ 2 N\geq2 N ≥ 2 —and, unlike for two-level emitters, also for odd N N N —there are interaction-dark sectors with active emitters (none is decoherence free: local pump, recycling, and decay move population between them). The sector-resolved feeding rate of Paper I, Φ n a , J = − g Im ⟨ a † S − Π n a , J ⟩ / ⟨ Π n a , J ⟩ \Phi_{n_a,J}=-g\,\operatorname{Im}\langle a^\dagger S_-\Pi_{n_a,J}\rangle/\langle\Pi_{n_a,J}\rangle Φ n a , J = − g Im ⟨ a † S − Π n a , J ⟩ / ⟨ Π n a , J ⟩ , connects the sectors to the output: by the first line of Eq. (8) ,
A ( ⟨ n ^ ⟩ − ν ) = ∑ n a , J ⟨ Π n a , J ⟩ Φ n a , J , (13) A(\langle\hat{n}\rangle-\nu)=\sum_{n_a,J}\langle\Pi_{n_a,J}\rangle\Phi_{n_a,J}, \tag*{(13)} A (⟨ n ^ ⟩ − ν ) = n a , J ∑ ⟨ Π n a , J ⟩ Φ n a , J , (13)
an exact decomposition of the emitted photon flux by sector, to which the J = 0 J=0 J = 0 sectors contribute nothing. Table IV applies it to two to five Λ \Lambda Λ emitters in the bad-cavity regime at P = A P=A P = A . The interaction-dark sectors with active emitters hold 10.8 % 10.8\% 10.8% , 9.7 % 9.7\% 9.7% , 8.3 % 8.3\% 8.3% , and 6.6 % 6.6\% 6.6% of the population for N = 2 , … , 5 N=2,\ldots,5 N = 2 , … , 5 , slowly decreasing with no even/odd alternation. The output is carried disproportionately by the high-pseudospin sectors, whose feeding rates grow faster than linearly in J J J and depend little on N N N , n a n_a n a , and P P P : at N = 2 N=2 N = 2 the bright triplet holds 38.5 % 38.5\% 38.5% of the population and emits 70 % 70\% 70% of the photons, and at N = 5 N=5 N = 5 the sectors with J ≥ 3 2 J\geq\frac{3}{2} J ≥ 2 3 hold 43 % 43\% 43% and emit 73%. How much more light the emitters would give without the dark sectors is a counterfactual that requires their residence times, which we have not computed.
Table IV. Sector-resolved stationary state of N = 2 , … , 5 N = 2,\ldots,5 N = 2 , … , 5 Λ \Lambda Λ emitters at g = 1.1 A g = 1.1A g = 1.1 A , γ = 0.3 A \gamma= 0.3A γ = 0.3 A , γ r = A \gamma_r = A γ r = A , P = A P = A P = A : dimension, population (%), feeding rate Φ n a , J / A \Phi_{n_a,J}/A Φ n a , J / A , and share of the output photon flux (%), Eq. (13) , of the sectors with n a n_a n a emitters in the reservoir and pseudo-spin J J J of the rest. Sectors with J = 0 J = 0 J = 0 are interaction dark. The flux shares add up to one to 2 × 10 − 15 2 \times10^{-15} 2 × 1 0 − 15 .
Scaling with N N N and the cumulant closure
Figure 9 shows ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ , g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , and C 2 ( N ) C_2^{(N)} C 2 ( N ) for N = 1 , … , 8 N = 1,\ldots,8 N = 1 , … , 8 in the bad-cavity regime at two pump rates, for the two-level, Λ \Lambda Λ , and ladder schemes, together with the second-order cumulant closure of Appendix C, the multilevel form of the closure benchmarked in Paper I. For all three schemes the photon number grows nearly linearly with N N N (per emitter: 0.11, 0.09, and a saturating 0.07 for the Λ \Lambda Λ , ladder, and two-level schemes at P = 0.3 A P = 0.3A P = 0.3 A ; 0.18, 0.14, 0.28 at P = A P = A P = A ). Single emitters are strongly antibunched (g ( 2 ) ( 0 ) = 0.32 g^{(2)}(0) = 0.32 g ( 2 ) ( 0 ) = 0.32 , 0.18, 0.55 at P = 0.3 A P = 0.3A P = 0.3 A ); pairs of Λ \Lambda Λ and ladder emitters are still sub-Poissonian (0.90, 0.83 at P = 0.3 A P = 0.3A P = 0.3 A ; 0.96, 0.87 at P = A P = A P = A ) where two-level pairs are already bunched (1.37, 1.11), and all three schemes approach g ( 2 ) ( 0 ) ≈ 1.24 g^{(2)}(0) \approx1.24 g ( 2 ) ( 0 ) ≈ 1.24 –1.26 by N = 8 N = 8 N = 8 at P = A P = A P = A , while the two-level laser at P = 0.3 A P = 0.3A P = 0.3 A heads towards thermal statistics (1.78). The pair coherence of the Λ \Lambda Λ and ladder media is positive and almost independent of N N N (+0.0030 to +0.0028 for Λ \Lambda Λ at P = 0.3 A P = 0.3A P = 0.3 A ); the two-level coherence is negative at P = 0.3 A P = 0.3A P = 0.3 A , decreasing in magnitude roughly as 1 / N 1/N 1/ N as in Paper I, and positive at P = A P = A P = A .
Figure 9. Scaling with emitter number at g = 1.1 A g = 1.1A g = 1.1 A , γ = 0.3 A \gamma= 0.3A γ = 0.3 A , γ r = γ p = A \gamma_r = \gamma_p = A γ r = γ p = A , for P = 0.3 A P = 0.3A P = 0.3 A (top) and P = A P = A P = A (bottom): (a,d) photon number, (b,e) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , (c,f) pair coherence, for the two-level, Λ \Lambda Λ , and ladder schemes; the inset of (c) shows the Λ \Lambda Λ and ladder coherences on their own scale. Symbols: exact continued fraction; dotted: second-order cumulant closure, Appendix C. The N = 8 N = 8 N = 8 values are reproducible under a change of radial cutoff from n max = 30 n_{\max} = 30 n m a x = 30 to 40 to better than 2 × 10 − 13 2 \times10^{-13} 2 × 1 0 − 13 (relative).
The closure is semi-quantitative for the photon number and poor for the emitter variables. It overestimates ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ of the Λ \Lambda Λ and ladder lasers by 13–15% at N = 1 N = 1 N = 1 rising to 15–17% at N = 8 N = 8 N = 8 at both pump rates (two-level: +7% falling to 0% at P = 0.3 A P = 0.3A P = 0.3 A , +14–18% at P = A P = A P = A ). It underestimates the laser-level inversion by a factor of two to three and a half for the Λ \Lambda Λ and ladder schemes at every N N N , already at N = 1 N = 1 N = 1 (0.036 against the exact 0.067 for one Λ \Lambda Λ emitter at P = 0.3 A P = 0.3A P = 0.3 A ), and by up to a factor of four for two-level emitters at P = A P = A P = A : the factorization ⟨ n ^ W u l ⟩ → ⟨ n ^ ⟩ ⟨ W u l ⟩ \langle\hat{n}W_{\rm ul}\rangle\to\langle\hat{n}\rangle\langle W_{\rm ul}\rangle ⟨ n ^ W ul ⟩ → ⟨ n ^ ⟩ ⟨ W ul ⟩ misses the anticorrelation of photon number and inversion identified in Paper I, which is stronger when the lower laser level is emptied by recycling. The sign of the pair coherence is correct in every case shown, but its magnitude is underestimated by up to 36% (P = 0.3 A P = 0.3A P = 0.3 A ) and 43% (P = A P = A P = A ) at N = 8 N = 8 N = 8 for the Λ \Lambda Λ and ladder media, and overestimated by 40% for two-level emitters at P = 0.3 A P = 0.3A P = 0.3 A . As in Paper I, the closure does not determine g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) and therefore cannot see the sub-Poissonian statistics of the multilevel pairs.
Counting statistics at larger N N N
Does the separation between equal-time and counting statistics found in Sec. VI persist as emitters are added? Figure 10 answers this at one benchmark point (g = 1.1 A g = 1.1A g = 1.1 A , γ = 0.1 A \gamma= 0.1A γ = 0.1 A , P = 0.5 A P = 0.5A P = 0.5 A , γ r = γ p = A \gamma_r = \gamma_p = A γ r = γ p = A ), using the norm-rescaled solver of Appendix E beyond the range of the plain recursion. It does: although g ( 2 ) ( 0 ) > 1 g^{(2)}(0) > 1 g ( 2 ) ( 0 ) > 1 for N ≥ 3 N \ge3 N ≥ 3 (with a flat maximum of 1.158 near N = 10 N = 10 N = 10 for Λ \Lambda Λ emitters), the long-time counting Fano factor stays below one and rises only slowly with N N N , from 0.39 at N = 1 N = 1 N = 1 to 0.74 at N = 12 N = 12 N = 12 for Λ \Lambda Λ emitters and from 0.34 to 0.55 at N = 8 N = 8 N = 8 for the ladder. The regularity imposed by the rate-limited cycle thus survives in the integrated output where the intracavity distribution is already super-Poissonian, and is eroded only gradually by the common cavity. These are finite-N N N results at fixed single-emitter coupling and fixed pump rate per emitter, so that the collective coupling grows as N \sqrt{N} N and the total injection as N N N ; they are not a large-N N N limit. The stationary observables at this point, and their reproducibility up to N = 16 N = 16 N = 16 , are listed in Table VII .
Figure 10. (a) g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) and (b) long-time counting Fano factor versus emitter number for Λ \Lambda Λ and ladder emitters at the benchmark point of Table VII (g = 1.1 A g = 1.1A g = 1.1 A , γ = 0.1 A \gamma= 0.1A γ = 0.1 A , P = 0.5 A P = 0.5A P = 0.5 A , γ r = γ p = A \gamma_r = \gamma_p = A γ r = γ p = A ; not the γ = 0.3 A \gamma= 0.3A γ = 0.3 A series of Fig. 9), from the norm-rescaled solver of Appendix E. Thin lines mark unity.
Table VII. TABLE VII. Block dimension, time of one stationary solve at n max = 30 n_{\max}=30 n m a x = 30 , and reproducibility, for Λ \Lambda Λ emitters (d = 3 d=3 d = 3 ) and ladder emitters (d = 4 d=4 d = 4 ) at g = 1.1 A g=1.1A g = 1.1 A , γ = 0.1 A \gamma=0.1A γ = 0.1 A , P = 0.5 A P=0.5A P = 0.5 A , γ r = γ p = A \gamma_r=\gamma_p=A γ r = γ p = A . “Plain”: the recursion in the damping basis used for Secs. IV–VIII, with the largest relative change of ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ and g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) between n max = 30 n_{\max}=30 n m a x = 30 and 40 (at N = 12 N=12 N = 12 it fails, see text). “Norm-rescaled”: the solver of Appendix § 4.4 , whose values are listed; “cutoff” is the largest relative change of ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ , g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , C 2 C_2 C 2 between n max = 30 n_{\max}=30 n m a x = 30 and 40 (ᵇbetween 20 and 30), “basis” the same against the fixed local basis (ᵃwith grade scaling λ = 0.3 \lambda=0.3 λ = 0.3 ). The spreads are reproducibility measures, not accuracy bounds.
Discussion and scope
a. Relation to other methods. Permutation-invariant Fock-space solvers [19 , 20 , 22 , 23 ] work in the same reduced emitter space when the pumping is incoherent [22 ] and give spectra and intensity correlations by quantum regression, as our independent check of the counting statistics does; the damping basis adds the block-Jacobi structure with its provably unique linear closure and convergence theorem, the moment functionals, the exactly terminating weak-pump series that produced Table II and Figure 2 , and the regression quantities within the same recurrence; the Fock solvers have no ⟨ n ^ ⟩ \langle\hat{n} \rangle ⟨ n ^ ⟩ conditioning limit and remain preferable above ⟨ n ^ ⟩ ≈ 20 \langle\hat{n} \rangle\approx20 ⟨ n ^ ⟩ ≈ 20 (Appendix § 4.3 ). Relative to the one-atom multilevel laser theory of Briegel, Meyer, and Englert [12 ] , which derives an effective pump operator for one atom, the present construction keeps the full collective structure of N N N emitters and the full level kinetics. The physical statements made here are statements about the specified cycles at the specified rates, except for Propositions 3 and 4, which hold for any independent incoherent rate graph satisfying their hypotheses; Proposition 4 is a sufficient condition for quenching, and the shelving example of § 3 shows that it is not necessary.
b. Limitations. The construction requires identical emitters with identical couplings and rates, one cavity-coupled transition, and incoherent pumping; coherent pump fields, as in Raman-pumped single-ion lasers [13 ] , enlarge the local space towards d 2 d^2 d 2 and change the grade bound (Remark 1). Within these restrictions, the differences between schemes found here matter where two or more incoherent steps of the excitation cycle have comparable rates, as for multistep carrier capture and relaxation in quantum-dot gain media [17 ] ; the model parameters used here are illustrative and not fitted to a specific device. The double-precision reach of the plain recursion in the damping basis is N ≲ 10 N \lesssim10 N ≲ 10 for three-level and N ≲ 8 N \lesssim8 N ≲ 8 for four-level emitters; the norm-rescaled coordinates of Appendix § 4.4 extend the tested range at the benchmark point (validated in detail at N = 12 N = 12 N = 12 , cross-checked through N = 16 N = 16 N = 16 for Λ \Lambda Λ and N = 10 N = 10 N = 10 for ladder emitters), but have not been applied to the parameter scans of Secs. IV–VI, and the e ⟨ n ^ ⟩ e^{\langle\hat{n} \rangle} e ⟨ n ^ ⟩ limit ⟨ n ^ ⟩ ≲ 20 \langle\hat{n} \rangle\lesssim20 ⟨ n ^ ⟩ ≲ 20 –25 of the damping basis remains (Appendix § 4.3 ). The weak-pump orders are statements about the pump series of the finite truncations (which have cutoff-independent coefficients); the direct stationary bound transfers only to schemes in which every charged level has a charge-lowering jump. The convergence results and Proposition 4 are for a zero-temperature cavity, the former under (U) and (P). The dark-sector populations are reported for N ≤ 5 N \le5 N ≤ 5 . The matched-photon-number pump-cycle comparisons of Sec. VI are for N ≤ 3 N \le3 N ≤ 3 ; the additional benchmark counting calculations of Sec. VIII reach N = 12 N = 12 N = 12 for Λ \Lambda Λ and N = 8 N = 8 N = 8 for ladder emitters, and the stationary low-order observables N = 16 N = 16 N = 16 and N = 10 N = 10 N = 10 , at one parameter point. The test of the effective two-level pump fits one parameter; two-level models with further fitted parameters were not examined. Open problems are the residence times of the dark sectors, the optimization of pump-cycle rates for noise suppression at given N N N , a controlled enlargement of the local space for coherent drives, and an analytic form of the sign boundary of Figure 2 .
c. Larger emitter numbers.* With the norm-rescaled coordinates the practical limitation is dense-block cost, O ( D N 3 ) O(D_N^3) O ( D N 3 ) per recursion step with D N = ( N + d + 1 d + 1 ) D_N = \binom{N+d+1}{d+1} D N = ( d + 1 N + d + 1 ) . Sparse or matrix-free solution of the stationary and Poisson problems, and, at weak pump, a basis restricted by the filtration of Proposition 3, are natural routes beyond the low tens; neither is implemented here.
Conclusion
The exact damping-basis continued fraction of Paper I extends to d d d -level emitters with arbitrary independent incoherent kinetics, in the reduced operator space of populations and lasing coherences, with the structural and convergence results intact under explicit hypotheses. The pump scheme is not a detail of few-emitter laser modelling: pumping out of a laser level forces self-quenching, while the Λ \Lambda Λ and ladder cycles studied here saturate to explicit limiting models and convert a larger fraction of the absorbed excitation current into cavity photons; the number of pump quanta needed to enter the laser manifold sets the order of the inter-emitter coherence, whose weak-pump sign is fixed by a recycling-rate boundary; rate-limited steps in the pump cycle lower the output-counting Fano factor of two and three emitters below one, a single-emitter effect that the common cavity erodes but that remains substantial in the counting statistics where the equal-time statistics hide it, because the anticorrelation forms at finite delay; a two-level pump fitted to the same photon number reproduces neither statistic for two and three emitters, nor the order and sign of the pair coherence, and becomes adequate where a single step of the cycle is rate limiting; and reservoir levels create dark sectors at every emitter number, which the exact sector decomposition of the output flux shows to emit nothing while the high-pseudo-spin sectors carry most of the light. Each of these is computed exactly at the few-emitter scales where cumulant closures, as the benchmark of Sec. VIII shows, are least reliable.
Data availability
The code and data that produce every number, table, and figure of this paper, including the regression tests and the independent Fock-space checks, are openly available in the Zenodo repository of Ref. [27 ] (doi:10.5281/zenodo.23841957, version 1.0). Its README maps every result to a script; the principal ones are, in the folder results/: Table II, Fig. 2, and the weak-pump sign boundary, fig2_signmap_weak_pump_series.py, with the finite-pump boundaries in checks_v3.py; Table III, pump_step_comparison.py; the independent-source comparison of Sec. VI, independent_source_reference.py; Fig. 8, fig8_g2_tau.py; the independent Fock-space check of F c o u n t F_{\mathrm{count}} F count , independent_counting_check.py; the larger-N N N results of Fig. 10 and Table VII, large_N_validation.py; and the stand-alone implementation of Appendix F, reconstruct_from_appendix.py.
Acknowledgments
Claude Fable 5 (Anthropic) was used as an assistive tool for literature searches, manuscript editing, code generation, and symbolic and numerical checks, including the preparation of plots. All AI-assisted code and calculations were independently checked, and all references and text were independently reviewed by the author, who takes full responsibility for the manuscript.
Table V. Atomic block dimension N L − 1 N^{L-1} N L − 1 for the reduced local space, L = d + 2 L=d+2 L = d + 2 , and for the full local Liouville space, L = d 2 L=d^{2} L = d 2 .
Construction details
a. Proof of Proposition 1. (i) For J = ∣ i ′ ⟩ ⟨ i ∣ J=|i'\rangle\langle i| J = ∣ i ′ ⟩ ⟨ i ∣ , D [ J ] P k = δ i k ( P i ′ − P i ) \mathcal{D}[J]P_k=\delta_{ik}(P_{i'}-P_i) D [ J ] P k = δ ik ( P i ′ − P i ) and D [ J ] τ + = J τ + J † − 1 2 { P i , τ + } = − 1 2 ( δ i u + δ i l ) τ + \mathcal{D}[J]\tau_+=J\tau_+J^\dagger-\frac{1}{2}\{P_i,\tau_+\}=-\frac{1}{2}(\delta_{iu}+\delta_{il})\tau_+ D [ J ] τ + = J τ + J † − 2 1 { P i , τ + } = − 2 1 ( δ i u + δ i l ) τ + , because ⟨ i ∣ u ⟩ ⟨ l ∣ i ⟩ = 0 \langle i|u\rangle\langle l|i\rangle=0 ⟨ i ∣ u ⟩ ⟨ l ∣ i ⟩ = 0 for every i i i ; likewise for τ − \tau_- τ − . D [ P u ] P k = 0 \mathcal{D}[P_u]P_k=0 D [ P u ] P k = 0 , D [ P u ] τ ± = − 1 2 τ ± \mathcal{D}[P_u]\tau_\pm=-\frac{1}{2}\tau_\pm D [ P u ] τ ± = − 2 1 τ ± , and the level Hamiltonian commutes with every P k P_k P k and gives [ P i , τ ± ] = ± ( δ i u − δ i l ) τ ± [P_i,\tau_\pm]=\pm(\delta_{iu}-\delta_{il})\tau_\pm [ P i , τ ± ] = ± ( δ i u − δ i l ) τ ± . (ii) τ + P k = δ k l τ + \tau_+P_k=\delta_{kl}\tau_+ τ + P k = δ k l τ + , P k τ + = δ k u τ + P_k\tau_+=\delta_{ku}\tau_+ P k τ + = δ k u τ + , τ + τ − = P u \tau_+\tau_-=P_u τ + τ − = P u , τ − τ + = P l \tau_-\tau_+=P_l τ − τ + = P l , τ ± 2 = 0 \tau_\pm^2=0 τ ± 2 = 0 , and the mirror identities. (iii) 1 = ∑ k P k \mathbb{1}=\sum_k P_k 1 = ∑ k P k . The Liouvillian is a sum of one-emitter Liouvillians, field terms, and the interaction, which acts on the emitters only by left and right multiplication with τ ± ( j ) \tau_\pm^{(j)} τ ± ( j ) ; so F ⊗ A ⊗ N \mathcal{F}\otimes\mathcal{A}^{\otimes N} F ⊗ A ⊗ N is invariant, and so is its permutation-invariant part. The maximally mixed emitter state lies in this closed subspace, and for a Fock truncation its evolution converges to the unique stationary state, which therefore lies in the subspace; the convergence of Fock truncations (Appendix B) transfers the statement to ρ s s \rho_{\mathrm{ss}} ρ ss . The dimension is the number of weak compositions of N N N into d + 2 d+2 d + 2 parts.
A coherent drive on any transition, or a second cavity-coupled transition, enlarges A \mathcal{A} A by the coherences it generates, up to the full d 2 d^2 d 2 ; Remark 1 says what survives.
b. Local bases. Two bases of A \mathcal{A} A are used in the physics sections, both with exactly one traceful element b 0 b_0 b 0 and traceless b 1 , … , b d − 1 b_1,\ldots,b_{d-1} b 1 , … , b d − 1 , so that the normalization and the observable functionals below are the same in either. The damping basis consists of r κ = ∑ i ( w κ ) i P i r_\kappa=\sum_i(w_\kappa)_iP_i r κ = ∑ i ( w κ ) i P i , with w 0 = p ˉ w_0=\bar p w 0 = p ˉ and w 1 , … , w d − 1 w_1,\ldots,w_{d-1} w 1 , … , w d − 1 the remaining right eigenvectors of W W W (traceless, because the left null vector of W W W is ( 1 , … , 1 ) (1,\ldots,1) ( 1 , … , 1 ) ), and r ± = τ ± r_\pm=\tau_\pm r ± = τ ± ; these are eigenoperators of the one-emitter Liouvillian with eigenvalues − β κ -\beta_\kappa − β κ (the eigenvalues of − W -W − W , complex for cyclic rate graphs) and − C u l ∓ i Ω -C_{ul}\mp i\Omega − C u l ∓ i Ω , Eq. (4) . The fixed basis consists of P g P_g P g for a reference level g g g , the traceless P i − P g P_i-P_g P i − P g (i ≠ g i\ne g i = g ), and τ ± \tau_\pm τ ± ; it is independent of all rates, the atomic part of the recurrence is not diagonal in it, and it is the basis in which the pump series of Sec. IV is constructed. The damping basis is used when W W W is diagonalizable with a well-conditioned eigenvector matrix, the fixed basis otherwise (defective W W W do occur, e.g. the Λ \Lambda Λ scheme at P = γ = γ r / 4 P=\gamma=\gamma_r/4 P = γ = γ r /4 , with eigenvalues 0 , − 3 γ r / 4 , − 3 γ r / 4 0,-3\gamma_r/4,-3\gamma_r/4 0 , − 3 γ r /4 , − 3 γ r /4 ); the implementation [27 ] switches automatically and the two give identical results where both apply (Appendix D).
The cavity damping basis ρ n ( k ) \rho_n^{(k)} ρ n ( k ) consists of the right eigenoperators of L f = A ( 1 + ν ) D [ a ] + A ν D [ a † ] − i ω [ n ^ , ⋅ ] \mathcal{L}_f=A(1+\nu)\mathcal{D}[a]+\mathcal{A}\nu\mathcal{D}[a^\dagger]-i\omega[\hat n,\cdot] L f = A ( 1 + ν ) D [ a ] + A ν D [ a † ] − iω [ n ^ , ⋅ ] ; left and right multiplication by a a a , a † a^\dagger a † change k k k by one and n n n by at most one through the coefficients f μ , δ L , R ( n , k ) f_{\mu,\delta}^{L,R}(n,k) f μ , δ L , R ( n , k ) of Table VIII , Appendix F. A third, rate-independent orthogonal basis is used for the largest emitter numbers (Appendix E).
Table VIII. Nonzero field coefficients f μ , δ L / R ( n , k ) f_{\mu,\delta}^{\mathrm{L/R}}(n,k) f μ , δ L/R ( n , k ) of Eq. (F2), with r = ν / ( 1 + ν ) r=\nu/(1+\nu) r = ν / ( 1 + ν ) ; all others vanish.
c. Proof of Lemma 1. Each of the N ( m ) \mathcal{N}(m) N ( m ) arrangements in R m \mathcal{R}_m R m has m μ m_\mu m μ sites carrying b μ b_\mu b μ , replacing one of them by b ν b_\nu b ν yields an arrangement of type m ′ = m − e μ + e ν m'=m-e_\mu+e_\nu m ′ = m − e μ + e ν , and each such arrangement is produced N ( m ) m μ / N ( m ′ ) = m ν ′ \mathcal{N}(m)m_\mu/\mathcal{N}(m')=m'_\nu N ( m ) m μ / N ( m ′ ) = m ν ′ times, so that the sum over sites and arrangements is x ν μ m μ N ( m ) R m ′ x_{\nu\mu}m_\mu\mathcal{N}(m)\mathcal{R}_{m'} x ν μ m μ N ( m ) R m ′ .
Applied to τ ± \tau_{\pm} τ ± , with τ + τ − = ( w l ) 1 τ + \tau_+\tau_-=(w_l)_1\tau_+ τ + τ − = ( w l ) 1 τ + and τ + τ − = P u = ∑ k c k ( u ) r k \tau_+\tau_-=P_u=\sum_k c_k^{(u)}r_k τ + τ − = P u = ∑ k c k ( u ) r k , the rule gives S + R m = ∑ k m k ( w k ) l R m − e k + e + + m − ∑ k c k ( u ) R m − e − + e k S_+R_m=\sum_k m_k(w_k)_lR_{m-e_k+e_+}+m_-\sum_k c_k^{(u)}R_{m-e_-+e_k} S + R m = ∑ k m k ( w k ) l R m − e k + e + + m − ∑ k c k ( u ) R m − e − + e k and three analogous formulas, which for d = 2 d=2 d = 2 reproduce Eqs. (5) –(8) of Paper I term by term; the same rule lifts the one-emitter Liouvillian. In the implementation the structure constants are obtained by projecting τ ± b μ \tau_\pm b_\mu τ ± b μ , b μ τ ± b_\mu\tau_\pm b μ τ ± , and L o c b μ \mathcal{L}_{\mathrm{oc}}b_\mu L oc b μ onto A \mathcal{A} A , the vanishing of the projection residual being the numerical form of Proposition 1.
Remark 1 (Scope). Items (i)–(ii) and the nilpotency in (iii) hold for any finite collection of local operators closed under the one-emitter Liouvillian and under multiplication by the cavity-coupled operators, provided a level charge exists that every term of S + S_+ S + raises by exactly one and that the coherent part of the dynamics conserves. The grade range, and with the nilpotency index and the length of the Neumann series, is then ∣ K ∣ + N ( h max + h min ) |K|+N(h_{\max}+h_{\min}) ∣ K ∣ + N ( h m a x + h m i n ) with h h h the spectral range of the local charge, not N + ∣ K ∣ N+|K| N + ∣ K ∣ : a three-level ladder with both adjacent transitions coupled to the same mode has charges 0 , 1 , 2 0,1,2 0 , 1 , 2 and index up to 2 N + ∣ K ∣ + 1 2N+|K|+1 2 N + ∣ K ∣ + 1 . We do not pursue such models here.
d. Numerical solution. For N ≤ 10 N\le10 N ≤ 10 we extract M ( 1 ) M^{(1)} M ( 1 ) , M ( 0 ) M^{(0)} M ( 0 ) , F ( 1 ) F^{(1)} F ( 1 ) , G G G by finite differences at a bulk index, set R n max = R ∞ \mathcal{R}_{n_{\max}}=\mathcal{R}_\infty R n m a x = R ∞ , recurse down to R 0 \mathcal{R}_0 R 0 , take X 0 X_0 X 0 as the right singular vector of M 0 + G R 0 M_0+GR_0 M 0 + G R 0 for its smallest singular value, normalize the component along Φ 0 , N ϵ 0 \Phi_{0,N\epsilon_0} Φ 0 , N ϵ 0 to one, and reconstruct X n + 1 = R n X n X_{n+1}=R_nX_n X n + 1 = R n X n ; for larger N N N the norm-rescaled form of Appendix E is used.
What the inherited proofs use
We list, for each item of Proposition 2, the properties of the emitter sector on which the corresponding proof in Paper I rests, and why they hold for (1). (i) The uncoupled atomic Liouvillian is diagonal (damping basis) or a general matrix (fixed basis) on the R m \mathcal{R}_m R m , which the recurrence does not need to be diagonal; the interaction changes n n n by at most one through the field identities, which are untouched; and K K K is conserved because τ + \tau_+ τ + raises q q q by exactly one on every nonvanishing term of lifted multiplication — τ + r κ ∝ τ + \tau_{+}r_{\kappa}\propto\tau_{+} τ + r κ ∝ τ + and τ + τ − ∈ span { r κ } \tau_{+}\tau_{-}\in\operatorname{span}\{r_{\kappa}\} τ + τ − ∈ span { r κ } — and τ − \tau_{-} τ − lowers it.
(ii) The proof of Proposition 2 of Paper I identifies M ( 1 ) M^{(1)} M ( 1 ) from the n n n -linear field coefficients alone and then uses only that [ S + , ⋅ ] [S_{+},\cdot] [ S + , ⋅ ] raises and [ S − , ⋅ ] [S_{-},\cdot] [ S − , ⋅ ] lowers the charge of every basis state by one, together with ∣ q − K ∣ ≤ N + ∣ K ∣ |q-K|\le N+|K| ∣ q − K ∣ ≤ N + ∣ K ∣ .
(iii) Proposition 3 of Paper I uses the field coefficients and the same charge property for F ( 1 ) F^{(1)} F ( 1 ) , N K \mathcal{N}_{K} N K , and G G G ; the statement for R ∞ R_{\infty} R ∞ is that each term of its Neumann series lowers the grade by j + 1 ≥ 1 j+1\ge1 j + 1 ≥ 1 , so that R ∞ R_{\infty} R ∞ lowers it by at least one (the proof of Proposition 3 of Paper I says “at least”; grade changes of − 1 -1 − 1 and − 2 -2 − 2 both occur, e.g. for two Λ \Lambda Λ emitters).
(iv) Proposition 4 of Paper I uses only the affine structure of the blocks, constant G G G , the invertibility of M ( 1 ) M^{(1)} M ( 1 ) , and, for the extension from n ≥ n 0 n\ge n_{0} n ≥ n 0 to all n n n , hypothesis (P), stated there in the same form.
(v) In the proof of Lemma 1 of Paper I, H H H commutes with E \mathcal{E} E because a † S − a^{\dagger}S_{-} a † S − and a S + aS_{+} a S + do; the jumps a a a and ∣ i ′ ⟩ ⟨ u ∣ |i'\rangle\langle u| ∣ i ′ ⟩ ⟨ u ∣ lower E \mathcal{E} E by one, the jumps ∣ u ⟩ ⟨ i ∣ |u\rangle\langle i| ∣ u ⟩ ⟨ i ∣ raise it by one at total rate ∑ j ∑ i ≠ u γ u ← i P i ( j ) ≤ Γ ↑ N \sum_{j}\sum_{i\ne u}\gamma_{u\leftarrow i}P_{i}^{(j)}\le\Gamma_{\uparrow}N ∑ j ∑ i = u γ u ← i P i ( j ) ≤ Γ ↑ N , and all other jumps preserve it, so L † e θ E = e θ E Q θ \mathcal{L}^{\dagger}e^{\theta\mathcal{E}}=e^{\theta\mathcal{E}}Q_{\theta} L † e θ E = e θ E Q θ (called W W W there) with Q θ ≤ ( e − θ − 1 ) A n ^ + ( e θ − 1 ) Γ ↑ N Q_{\theta}\le(e^{-\theta}-1)A\hat n+(e^{\theta}-1)\Gamma_{\uparrow}N Q θ ≤ ( e − θ − 1 ) A n ^ + ( e θ − 1 ) Γ ↑ N at ν = 0 \nu=0 ν = 0 , and the rest of the proof is unchanged. The proof of Proposition 5 of Paper I uses this bound, the compactness of the resolvent of E \mathcal{E} E , and (U); the proof of Theorem 1 (Appendix C of Paper I) uses the bound, item (iv), the grade bound ∣ k α ∣ ≤ N |k_{\alpha}|\le N ∣ k α ∣ ≤ N , the U ( 1 ) U(1) U ( 1 ) covariance of the master equation — which holds for eq:1 under ∣ u ⟩ → e i φ ∣ u ⟩ |u\rangle\to e^{i\varphi}|u\rangle ∣ u ⟩ → e i φ ∣ u ⟩ , a → e i φ a a\to e^{i\varphi}a a → e i φ a , since D [ ∣ i ′ ⟩ ⟨ i ∣ ] \mathcal{D}[|i'\rangle\langle i|] D [ ∣ i ′ ⟩ ⟨ i ∣ ] is invariant under any rephasing of the levels — and permutation invariance. None of these refer to d = 2 d=2 d = 2 . The normalization convention (component along Φ 0 , N e 0 \Phi_{0,Ne_{0}} Φ 0 , N e 0 equal to one) requires a local basis with a single traceful element, which both bases of Sec. III provide.
Second-order cumulant closure for multilevel emitters
The exact equations of motion for ⟨ n ^ ⟩ \langle\hat n\rangle ⟨ n ^ ⟩ , the one-emitter level populations p i = ⟨ P i ( j ) ⟩ p_i=\langle P_i^{(j)}\rangle p i = ⟨ P i ( j ) ⟩ , Z 1 = ⟨ a † τ − ( j ) ⟩ Z_1=\langle a^{\dagger}\tau_{-}^{(j)}\rangle Z 1 = ⟨ a † τ − ( j ) ⟩ , and C 2 = ⟨ τ + ( i ) τ − ( j ) ⟩ C_2=\langle\tau_{+}^{(i)}\tau_{-}^{(j)}\rangle C 2 = ⟨ τ + ( i ) τ − ( j ) ⟩ (i ≠ j i\ne j i = j ) couple to the third-order moments ⟨ a † a ( P u − P l ) ( j ) ⟩ \langle a^{\dagger}a(P_u-P_l)^{(j)}\rangle ⟨ a † a ( P u − P l ) ( j ) ⟩ and ⟨ a † ( P u − P l ) ( i ) τ − ( j ) ⟩ \langle a^{\dagger}(P_u-P_l)^{(i)}\tau_{-}^{(j)}\rangle ⟨ a † ( P u − P l ) ( i ) τ − ( j ) ⟩ . Setting the corresponding third-order cumulants to zero, as in eq:33 of Paper I (the first moments of the charged operators vanish by U ( 1 ) U(1) U ( 1 ) symmetry), gives, at ν = 0 \nu=0 ν = 0 and with w = p u − p l w=p_u-p_l w = p u − p l ,
d d t ⟨ n ^ ⟩ = − A ⟨ n ^ ⟩ − g N Im Z 1 , d d t p = W p + g Im Z 1 ( e u − e l ) , d d t Z 1 = − ( A / 2 + C u l + i Δ ) Z 1 − i g 2 [ p u + ( N − 1 ) C 2 + ⟨ n ^ ⟩ w ] , d d t C 2 = − 2 C u l C 2 − g w Im Z 1 , (14) \begin{aligned}
\frac{d}{dt}\langle\hat n\rangle&= -A\langle\hat n\rangle-gN\operatorname{Im}Z_1,\\
\frac{d}{dt}p &= Wp+g\operatorname{Im}Z_1(e_u-e_l),\\
\frac{d}{dt}Z_1 &= -(A/2+C_{ul}+i\Delta)Z_1-\frac{ig}{2}\left[p_u+(N-1)C_2+\langle\hat n\rangle w\right],\\
\frac{d}{dt}C_2 &= -2C_{ul}C_2-gw\operatorname{Im}Z_1, \tag*{(14)}
\end{aligned} d t d ⟨ n ^ ⟩ d t d p d t d Z 1 d t d C 2 = − A ⟨ n ^ ⟩ − g N Im Z 1 , = W p + g Im Z 1 ( e u − e l ) , = − ( A /2 + C u l + i Δ ) Z 1 − 2 i g [ p u + ( N − 1 ) C 2 + ⟨ n ^ ⟩ w ] , = − 2 C u l C 2 − g w Im Z 1 , (14)
with W W W the rate matrix (3). For d = 2 d=2 d = 2 these are the closure equations of Paper I, and the implementation reproduces its quoted closure values (C 2 = − 0.017 C_2=-0.017 C 2 = − 0.017 , ⟨ σ z ⟩ c l = − 0.039 \langle\sigma_z\rangle_{\mathrm{cl}}=-0.039 ⟨ σ z ⟩ cl = − 0.039 at N = 3 N=3 N = 3 , s = 0.9 s=0.9 s = 0.9 , baseline parameters). The physical root is obtained by integrating (14) from the vacuum with all emitters in the stationary distribution of W W W . The closure contains no independent fourth-order field moment and does not determine g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) .
Validation, cost, conditioning
Validation
(a) With d = 2 d=2 d = 2 the implementation [27 ] reproduces the two-level code of Paper I: the weak-pump reference value ⟨ n ^ ⟩ / s = 0.34521131203 \langle\hat n\rangle/s=0.34521131203 ⟨ n ^ ⟩ / s = 0.34521131203 of its Appendix D2 to 10 − 11 10^{-11} 1 0 − 11 (relative) and the dominant spectral poles of its Table III to all nine digits shown (both are regression tests of the package). (b) For the V, Λ \Lambda Λ , and cyclic schemes (with pure dephasing), and ladder schemes at N = 1 , 2 N=1,2 N = 1 , 2 , ( Δ , ν ) ∈ { ( 0 , 0 ) , ( 0.4 A , 0.1 ) } (\Delta,\nu)\in\{(0,0),(0.4A,0.1)\} ( Δ , ν ) ∈ {( 0 , 0 ) , ( 0.4 A , 0.1 )} , and for the cyclic Λ \Lambda Λ scheme at N = 3 N=3 N = 3 , the photon number, g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , all level populations, Z Z Z , the pair coherence, and every element of the reconstructed reduced atomic state agree with a brute-force steady-state solution of the full d N × n c a v d^{N}\times n_{\mathrm{cav}} d N × n cav master equation (no symmetry reduction, n c a v = 16 n_{\mathrm{cav}}=16 n cav = 16 , or 12 at N = 3 N=3 N = 3 ) to 10 − 15 10^{-15} 1 0 − 15 –10 − 13 10^{-13} 1 0 − 13 at ν = 0 \nu=0 ν = 0 and to 5 × 10 − 11 5\times10^{-11} 5 × 1 0 − 11 at ν = 0.1 \nu=0.1 ν = 0.1 , the latter limited by the brute-force cutoff. (c) The largest-photon-number point of the scan of Figure 3 (N = 3 N=3 N = 3 , two-level, P / A = 32 P/A=32 P / A = 32 ) agrees with the brute-force solution at Fock cutoffs 70 and 85 to all eight digits of ⟨ n ^ ⟩ = 24.925794 \langle\hat n\rangle=24.925794 ⟨ n ^ ⟩ = 24.925794 , g ( 2 ) ( 0 ) = 1.0114823 g^{(2)}(0)=1.0114823 g ( 2 ) ( 0 ) = 1.0114823 , and C 2 = 0.10702979 C_2=0.10702979 C 2 = 0.10702979 , so that the representation limit discussed below does not affect it. (d) At the defective-W W W point P = γ = γ r / 4 P=\gamma=\gamma_r/4 P = γ = γ r /4 of the Λ \Lambda Λ scheme the automatic fallback to the fixed basis reproduces the raw-projector basis and the brute force to 10 − 15 10^{-15} 1 0 − 15 at N = 1 , 2 N=1,2 N = 1 , 2 , and the damping, fixed, and raw bases agree to 2 × 10 − 14 2\times10^{-14} 2 × 1 0 − 14 at neighboring points (recorded in the validation output; the regression tests assert 10 − 12 10^{-12} 1 0 − 12 and 10 − 11 10^{-11} 1 0 − 11 ). (e) The structural statements of Proposition 2 were checked for all four schemes at N ≤ 3 N\le3 N ≤ 3 , K ∈ { 0 , − 1 , 2 } K\in\{0,-1,2\} K ∈ { 0 , − 1 , 2 } , ν = 0.1 \nu=0.1 ν = 0.1 , Δ = 0.3 A \Delta=0.3A Δ = 0.3 A : the deviations of the explicitly constructed M n M_n M n , G n G_n G n , F n F_n F n at n = 0 , 1 , 13 n=0,1,13 n = 0 , 1 , 13 from their affine forms are at most 3 × 10 − 13 3\times10^{-13} 3 × 1 0 − 13 , every eigenvalue of N K \mathcal{N}_K N K vanishes to 3 × 10 − 15 3\times10^{-15} 3 × 1 0 − 15 , the nilpotency index of N K \mathcal{N}_K N K and of R ∞ R_{\infty} R ∞ equals N + ∣ K ∣ + 1 N+|K|+1 N + ∣ K ∣ + 1 exactly at K = 0 K=0 K = 0 and K = − 1 K=-1 K = − 1 in every case, and every nonzero element of F ( 1 ) F^{(1)} F ( 1 ) , N K \mathcal{N}_K N K , and G G G connects grades differing by exactly one in the stated direction. (f) The V scheme converges to the two-level model as γ r → ∞ \gamma_r\to\infty γ r → ∞ at the expected rate (Table VI): at N = 2 N=2 N = 2 , P = 8 A P=8A P = 8 A , the differences in all four observables decrease as 1 / γ r 1/\gamma_r 1/ γ r , which is the quantitative form of the statement that the two-level incoherent pump is the fast-reservoir limit of the V scheme, including its pump-induced broadening C ′ = ( γ + P ) / 2 C'=(\gamma+P)/2 C ′ = ( γ + P ) /2 . (g) The dominant pole of the regression spectrum is selected as the pole with the largest real part among those whose residue modulus (a conjugate pair counting jointly) exceeds a threshold; for all schemes of Figure 3 at P / A = 3 , 16 , 64 P/A=3,16,64 P / A = 3 , 16 , 64 the selected width is identical for thresholds of 1%, 5%, and 20%. (h) The counting Fano factor was recomputed independently in the full Fock⊗ \otimes ⊗ emitter space for five cases of Table III (two-level N = 1 N=1 N = 1 ; Λ \Lambda Λ N = 2 N=2 N = 2 ; ladder N = 2 N=2 N = 2 at ⟨ n ^ ⟩ = 0.3 \langle\hat{n}\rangle=0.3 ⟨ n ^ ⟩ = 0.3 and 1; ladder N = 3 N=3 N = 3 ), by building the Liouvillian with QuTiP and solving the Poisson equation for the connected regression integral; at Fock cutoffs 10–13 (11 at N = 3 N=3 N = 3 ) the results agree with the continued fraction to 6 × 10 − 15 6 \times10^{-15} 6 × 1 0 − 15 , 2 × 10 − 14 2 \times10^{-14} 2 × 1 0 − 14 , 2 × 10 − 13 2 \times10^{-13} 2 × 1 0 − 13 , 9 × 10 − 13 9 \times10^{-13} 9 × 1 0 − 13 , and 5 × 10 − 13 5 \times10^{-13} 5 × 1 0 − 13 . (i) Strong-pump limits and finite-pump boundaries (Secs. IV and § 3 ; data in results/checks v2 and results/checks v3 of Ref. [27 ] ). At N = 2 N=2 N = 2 , g = 5 A g=5A g = 5 A , γ = A \gamma=A γ = A , the Λ \Lambda Λ laser with γ r = 10 A \gamma_r=10A γ r = 10 A has ⟨ n ^ ⟩ = 6.167 \langle\hat{n}\rangle=6.167 ⟨ n ^ ⟩ = 6.167 , 6.623, 6.672, 6.677 at P / A = 64 P/A=64 P / A = 64 , 640, 6400, 64000 (limit 6.678), the fast ladder 4.113, 4.318, 4.339, 4.342 (V limit 4.342), the two-level laser 0.084, 0.0079, 0.00078 and the V laser 0.080, 0.0075, 0.00074 at P / A = 640 P/A=640 P / A = 640 , 6400, 64000; the shelving cycle has ⟨ n ^ ⟩ = 0.0675 \langle\hat{n}\rangle=0.0675 ⟨ n ^ ⟩ = 0.0675 , 0.0198, 0.00245, 0.000251 at P / A = 1 P/A=1 P / A = 1 , 10, 100, 1000. The pair coherence at P = 16 A P=16A P = 16 A vanishes at γ r / A = 3.545 \gamma_r/A=3.545 γ r / A = 3.545 (Λ \Lambda Λ ) and 3.511 (ladder) for N = 2 N=2 N = 2 and at 3.253 and 3.243 for N = 3 N=3 N = 3 , against 3.547 for both schemes and both N N N at P = 10 − 3 A P=10^{-3}A P = 1 0 − 3 A and the weak-pump series value 3.547. (j) The norm-rescaled solver used for N > 10 N>10 N > 10 and its validation are described in Appendix E. Checks (b), (c), and (h) are the comparisons with independent solvers; the series coefficients, sign map, spectra, and N ≤ 8 N \le8 N ≤ 8 results are validated through the structural checks and internal consistency, not individually by brute force.
Table VI. Convergence of the V scheme to the two-level incoherent-pump model as the reservoir decay γ r → ∞ \gamma_r \to\infty γ r → ∞ : differences of the photon number, g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , pair coherence, and dominant-pole width Γ 1 = − 2 Re λ 1 \Gamma_1=-2\operatorname{Re}\lambda_1 Γ 1 = − 2 Re λ 1 from the two-level values at N = 2 N=2 N = 2 , g = 5 A g=5A g = 5 A , γ = A \gamma=A γ = A , P = 8 A P=8A P = 8 A (two-level: ⟨ n ^ ⟩ = 5.4665 \langle\hat{n}\rangle=5.4665 ⟨ n ^ ⟩ = 5.4665 , g ( 2 ) ( 0 ) = 1.0447 g^{(2)}(0)=1.0447 g ( 2 ) ( 0 ) = 1.0447 , C 2 = 0.0403 C_2=0.0403 C 2 = 0.0403 , Γ 1 = 0.303 A \Gamma_1=0.303A Γ 1 = 0.303 A ).
b. Cost. The blocks are affine in n n n and are generated from four stored D N × D N D_N \times D_N D N × D N matrices. The plain stationary solver stores all transfer matrices R n R_n R n for the upward reconstruction, so its memory is O ( n max D N 2 ) O(n_{\max}D_N^2) O ( n m a x D N 2 ) : 0.8 GB for eight ladder emitters (D = 1287 D=1287 D = 1287 , n max = 30 n_{\max}=30 n m a x = 30 ). The norm-rescaled solver of Appendix E streams the recursion and keeps only R 0 R_0 R 0 and R 1 R_1 R 1 , so its memory is O ( D N 2 ) O(D_N^2) O ( D N 2 ) (a few dense blocks, about 3 GB at D = 4845 D=4845 D = 4845 ). The time of both is O ( n max D N 3 ) O(n_{\max}D_N^3) O ( n m a x D N 3 ) . The spectrum and the intensity correlation diagonalize a dense matrix of dimension ( n max + 1 ) D N (n_{\max}+1)D_N ( n m a x + 1 ) D N ; the block Poisson solver of Appendix E gives the counting Fano factor without an eigendecomposition at the cost of one stationary solve with stored R n R_n R n . The dense routes are reported separately; they dominate the cost of Figure 3 and Figure 6 (about four minutes for the N = 3 N=3 N = 3 ladder curve at 16 pump values). Table VII lists stationary-solve timings on a two-core virtual machine (Intel Xeon, 2.8 GHz) with numpy 2.5 linked to OpenBLAS running two threads, for both solvers.
c. Conditioning. The precision limits are those of Paper I: the e ⟨ n ^ ⟩ e^{\langle\hat{n}\rangle} e ⟨ n ^ ⟩ representation limit of the damping basis in double precision — which check (c) shows not to affect the ⟨ n ^ ⟩ ≈ 3 \langle\hat{n}\rangle\approx3 ⟨ n ^ ⟩ ≈ 3 point at N = 3 N=3 N = 3 , where the blocks are small — and the N N N -dependent conditioning of the pivots, which comes from the non-normality of the nilpotent part of M ( 1 ) M^{(1)} M ( 1 ) (index N + 1 N+1 N + 1 , as before) and sets in at smaller N N N because the blocks are larger. Table VII shows the plain double-precision recursion in the damping basis is reproducible under a change of radial cutoff to $10^{-12}$ (6 × 10 − 10 6 \times10^{-10} 6 × 1 0 − 10 for C 2 C_2 C 2 ) for eight ladder emitters, to 10 − 10 10^{-10} 1 0 − 10 (5 × 10 − 10 5 \times10^{-10} 5 × 1 0 − 10 for C 2 C_2 C 2 ) for eight Λ \Lambda Λ emitters, and to 3 × 10 − 7 3 \times10^{-7} 3 × 1 0 − 7 (10 − 6 10^{-6} 1 0 − 6 for C 2 C_2 C 2 ) for ten; for twelve Λ \Lambda Λ emitters ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ and g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) change by 0.7% and 0.8% between n max = 30 n_{\max}=30 n m a x = 30 and 40, the pair coherence is not reproducible at all, and the photon number is 20% off, because in these coordinates the stationary kernel of M 0 + G R 0 M_0+GR_0 M 0 + G R 0 is not separated from the next singular values (Appendix E). The practical reach of the plain recursion is thus N ≲ 10 N \lesssim10 N ≲ 10 for three-level and N ≲ 8 N \lesssim8 N ≲ 8 for four-level emitters. The norm-rescaled solver extends this range at the benchmark point: its results are reproducible under a change of radial cutoff to 10 − 13 10^{-13} 1 0 − 13 (relative) for D N ≤ 2000 D_N \le2000 D N ≤ 2000 (n max = 30 n_{\max}=30 n m a x = 30 versus 40) and to 2 × 10 − 12 2 \times10^{-12} 2 × 1 0 − 12 and 3 × 10 − 11 3 \times10^{-11} 3 × 1 0 − 11 at N = 14 N=14 N = 14 and 16 (20 versus 30); they agree with the fixed local basis to 10 − 12 10^{-12} 1 0 − 12 for ladder emitters through N = 10 N=10 N = 10 , and with the grade-scaled fixed basis (λ = 0.3 \lambda=0.3 λ = 0.3 ) to 4 × 10 − 10 4 \times10^{-10} 4 × 1 0 − 10 at N = 14 N=14 N = 14 and 2 × 10 − 10 2 \times10^{-10} 2 × 1 0 − 10 at $N=16 for Λ \Lambda Λ emitters (without grade scaling the fixed basis itself degrades with N N N , to 4 × 10 − 6 4 \times10^{-6} 4 × 1 0 − 6 at N = 16 N=16 N = 16 ); at N = 12 N=12 N = 12 they agree with an extended-precision recursion to 3 × 10 − 14 3 \times10^{-14} 3 × 1 0 − 14 . These are reproducibility measures across independent representations, not rigorous error bounds. Where the full local space would be needed (coherent drives) the reach at comparable block sizes is N ≤ 5 N \le5 N ≤ 5 for d = 3 d=3 d = 3 and N ≤ 3 N \le3 N ≤ 3 for d = 4 d=4 d = 4 .
Norm-rescaled solver for larger N N N
a. Why the plain recursion fails. The reach of the plain recursion (Table VII , “plain” columns) is limited by the coordinates, not by the equations. In the damping basis the three smallest singular values of M 0 + G R 0 M_0+GR_0 M 0 + G R 0 at N = 12 N=12 N = 12 (Λ \Lambda Λ benchmark of Table VII ) are 1.7 × 10 − 7 1.7 \times10^{-7} 1.7 × 1 0 − 7 , 3.0 × 10 − 7 3.0 \times10^{-7} 3.0 × 1 0 − 7 , 3.6 × 10 − 7 3.6 \times10^{-7} 3.6 × 1 0 − 7 : the stationary kernel is not separated, and the singular vector used as kernel gives a photon number 20% below the value obtained below.
b. Coordinates and blocks. The norm-rescaled solver (implementation [27 ] , core/normalized.py) solves the same recurrence after three exact changes of representation. (1) The local basis is b 0 = 1 / d b_0=\mathbb{1}/d b 0 = 1/ d , b 1 , … , b d − 1 b_1,\ldots,b_{d-1} b 1 , … , b d − 1 Hilbert–Schmidt orthonormal traceless diagonal operators, and b ± = τ ± b_{\pm}=\tau_{\pm} b ± = τ ± . (2) The collective basis is rescaled to unit norm: for mutually orthogonal b μ b_{\mu} b μ , ∥ R m ∥ H S 2 = ∏ μ ∥ b μ ∥ 2 m μ / N ( m ) \lVert R_m\rVert_{\mathrm{HS}}^2=\prod_{\mu}\lVert b_{\mu}\rVert^{2m_{\mu}}/\mathcal{N}(m) ∥ R m ∥ HS 2 = ∏ μ ∥ b μ ∥ 2 m μ / N ( m ) , and the new coordinates are x m = X m ∥ R m ∥ H S x_m=X_m\lVert R_m\rVert_{\mathrm{HS}} x m = X m ∥ R m ∥ HS , so that all blocks undergo the diagonal similarity B → W − 1 B W B\to W^{-1}BW B → W − 1 B W , W = diag ( ∥ R m ∥ H S − 1 ) W=\operatorname{diag}(\lVert R_m\rVert_{\mathrm{HS}}^{-1}) W = diag (∥ R m ∥ HS − 1 ) . The normalization and observable prescriptions of Sec. III must be transformed accordingly. Only R N e 0 = ( 1 / d ) ⊗ N R_{N e_0}=(\mathbb{1}/d)^{\otimes N} R N e 0 = ( 1/ d ) ⊗ N is traceful, with tr R N e 0 = 1 \operatorname{tr}R_{N e_0}=1 tr R N e 0 = 1 and ∥ R N e 0 ∥ H S = d − N / 2 \lVert R_{N e_0}\rVert_{\mathrm{HS}}=d^{-N/2} ∥ R N e 0 ∥ HS = d − N /2 ; the trace condition X 0 , N e 0 = 1 X_{0,N e_0}=1 X 0 , N e 0 = 1 therefore becomes x 0 , N e 0 = d − N / 2 x_{0,N e_0}=d^{-N/2} x 0 , N e 0 = d − N /2 , not one, and an observable functional v m = tr ( O R m ) v_m=\operatorname{tr}(O R_m) v m = tr ( O R m ) becomes v m / ∥ R m ∥ H S v_m/\lVert R_m\rVert_{\mathrm{HS}} v m / ∥ R m ∥ HS . (An overall constant factor in W W W is immaterial; the implementation fixes it for numerical range and converts back to X m X_m X m before evaluating observables.) (3) The affine coefficients M ( 1 ) M^{(1)} M ( 1 ) , M ( 0 ) M^{(0)} M ( 0 ) , G G G , F ( 1 ) F^{(1)} F ( 1 ) are assembled from the scalar field coefficients and sparse lifted one-site maps, rather than by differencing blocks at two radial indices, so that G ( 1 ) = F ( 0 ) = 0 G^{(1)}=F^{(0)}=0 G ( 1 ) = F ( 0 ) = 0 hold exactly rather than to rounding.
Algorithms
The downward recursion is streamed and keeps only R 0 R_0 R 0 and R 1 R_1 R 1 , which suffice for X 0 , X 1 , X 2 X_0,X_1,X_2 X 0 , X 1 , X 2 and hence for ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ , g ( 2 ) ( 0 ) g^{(2)}(0) g ( 2 ) ( 0 ) , the populations, and C 2 C_2 C 2 (memory O ( D N 2 ) O(D_N^2) O ( D N 2 ) ). The kernel is obtained by replacing one linearly dependent row of M 0 + G R 0 M_0+GR_0 M 0 + G R 0 (the row of the traceful element, which carries the trace functional as a left null vector) by the trace condition, instead of a singular-value decomposition. The counting Fano factor is obtained in the same coordinates from the Poisson equation of Sec. VI, M Z = − ( Y − ⟨ n ^ ⟩ X s s ) \mathcal{M}Z=-(Y-\langle\hat{n}\rangle X_{\mathrm{ss}}) M Z = − ( Y − ⟨ n ^ ⟩ X ss ) , by the block analogue of the continued fraction: Z n + 1 = R n Z n + t n + 1 Z_{n+1}=R_nZ_n+t_{n+1} Z n + 1 = R n Z n + t n + 1 with the stationary R n R_n R n , t n max = 0 t_{n_{\max}}=0 t n m a x = 0 , t n = ( M n + G R n ) − 1 ( b n − G t n + 1 ) t_n=(M_n+GR_n)^{-1}(b_n-Gt_{n+1}) t n = ( M n + G R n ) − 1 ( b n − G t n + 1 ) , and at n = 0 n=0 n = 0 the same row replacement with Z 0 = 0 Z_0=0 Z 0 = 0 . Only Z 0 , Z 1 Z_0,Z_1 Z 0 , Z 1 enter the integral, but the source needs all X n X_n X n , so the R n R_n R n are stored (memory O ( n max D N 2 ) O(n_{\max}D_N^2) O ( n m a x D N 2 ) ); no eigendecomposition of the ( n max + 1 ) D N (n_{\max}+1)D_N ( n m a x + 1 ) D N -dimensional matrix is needed.
Validation
At N = 12 N=12 N = 12 the solver gives ⟨ n ^ ⟩ = 2.377788932 \langle\hat{n}\rangle=2.377788932 ⟨ n ^ ⟩ = 2.377788932 , g ( 2 ) ( 0 ) = 1.1558613090 g^{(2)}(0)=1.1558613090 g ( 2 ) ( 0 ) = 1.1558613090 , C 2 = 0.0093404550 C_2=0.0093404550 C 2 = 0.0093404550 . These values are reproduced under four variations of the numerical representation (not independent physical solvers): radial cutoffs 20, 30, 40 (relative spread 2 × 10 − 14 2\times10^{-14} 2 × 1 0 − 14 , 10 − 13 10^{-13} 1 0 − 13 for C 2 C_2 C 2 ); additional grade rescalings λ ∣ k ∣ \lambda^{|k|} λ ∣ k ∣ , λ = 0.3 \lambda=0.3 λ = 0.3 and 3 (10 − 13 10^{-13} 1 0 − 13 and 5 × 10 − 11 5\times10^{-11} 5 × 1 0 − 11 ); the fixed local basis (P r e f , P i − P r e f P_{\mathrm{ref}},P_i-P_{\mathrm{ref}} P ref , P i − P ref ), non-orthogonal and related to the first by a dense transformation of every block (6 × 10 − 11 6\times10^{-11} 6 × 1 0 − 11 , 10 − 12 10^{-12} 1 0 − 12 with λ = 0.3 \lambda=0.3 λ = 0.3 ); and the recursion and kernel solve repeated in extended precision (64-bit mantissa, Gaussian elimination with partial pivoting) on the same double-precision blocks (5 × 10 − 15 5\times10^{-15} 5 × 1 0 − 15 ). The last check tests the propagation of rounding through the elimination, not the assembly of the blocks. Consistency diagnostics of the solution are recorded in the repository: at N = 12 , 14 , 16 N=12,14,16 N = 12 , 14 , 16 (Λ \Lambda Λ ) and N = 8 , 10 N=8,10 N = 8 , 10 (ladder) the normwise relative residual of the n = 0 n=0 n = 0 and n = 1 n=1 n = 1 block equations is below 3 × 10 − 18 3\times10^{-18} 3 × 1 0 − 18 (of the full truncated system at N = 12 N=12 N = 12 and 8 likewise), the trace condition and the sum of the level populations hold to 2 × 10 − 16 2\times10^{-16} 2 × 1 0 − 16 , all populations lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] , the imaginary parts of ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ , ⟨ n ^ 2 ⟩ \langle\hat{n}^2\rangle ⟨ n ^ 2 ⟩ , and C 2 C_2 C 2 vanish, and the Λ \Lambda Λ populations satisfy the stationary flux balance P p r = γ r p l Pp_r=\gamma_rp_l P p r = γ r p l to 5 × 10 − 16 5\times10^{-16} 5 × 1 0 − 16 . Small residuals do not by themselves bound the error of an ill-conditioned problem, which is why the variations above are the primary evidence. The counting solver reproduces the dense Poisson solve and the independent Fock value 0.7399491367 of Appendix § 4.3 , check (h), to 2 × 10 − 14 2\times10^{-14} 2 × 1 0 − 14 , is cutoff independent to 4 × 10 − 15 4\times10^{-15} 4 × 1 0 − 15 up to N = 10 N=10 N = 10 , and agrees with the fixed basis to 7 × 10 − 12 7\times10^{-12} 7 × 1 0 − 12 at N = 12 N=12 N = 12 . Results of the same variations through N = 16 N=16 N = 16 (Λ \Lambda Λ ) and N = 10 N=10 N = 10 (ladder) are given in Table VII and Appendix § 4.3 (“Conditioning”).
All spreads are reproducibility measures, not accuracy bounds.
e. Ablation. Relative deviation of ⟨ n ^ ⟩ \langle\hat{n}\rangle ⟨ n ^ ⟩ at N = 12 N = 12 N = 12 , n max = 30 n_{\max} = 30 n m a x = 30 , from the reference calculation (orthogonal basis, norm rescaling, algebraic blocks): plain solver as in Table VII , 0.2; same blocks with the trace-row kernel, 1 × 10 − 4 1 \times10^{-4} 1 × 1 0 − 4 ; trace-row kernel without norm rescaling, 3 × 10 − 4 3 \times10^{-4} 3 × 1 0 − 4 (damping basis), 2 × 10 − 9 2 \times10^{-9} 2 × 1 0 − 9 (fixed), and 2 × 10 − 12 2 \times10^{-12} 2 × 1 0 − 12 (orthogonal); with norm rescaling but differenced blocks, 4 × 10 − 14 4 \times10^{-14} 4 × 1 0 − 14 (orthogonal). The kernel extraction and the local basis therefore matter most; the norm rescaling contributes the last two orders. The damping basis is not rescued by the norm rescaling (5 × 10 − 5 5 \times10^{-5} 5 × 1 0 − 5 , and 3% with λ = 3 \lambda= 3 λ = 3 ): its eigenvector populations are far from orthogonal.
Assembly of the recurrence
This appendix lists everything needed to implement the recurrence of Sec. III without Paper I. A stand-alone implementation that follows it line by line and imports nothing from the package, results/reconstruct_from_appendix.py of Ref. [27 ] , reproduces the package to 2 × 10 − 14 2 \times10^{-14} 2 × 1 0 − 14 (relative) for the N = 2 N = 2 N = 2 rows of Table III at ⟨ n ^ ⟩ = 1 \langle\hat{n}\rangle= 1 ⟨ n ^ ⟩ = 1 and for ladder emitters at ν = 0.1 \nu= 0.1 ν = 0.1 , Δ = 0.4 A \Delta= 0.4A Δ = 0.4 A . We work in the frame rotating at the cavity frequency ω \omega ω , in which the term − i ω K -i\omega K − iω K of Sec. III B is absent.
a. Cavity damping basis. With k ± = max ( ± k , 0 ) k_{\pm} = \max(\pm k,0) k ± = max ( ± k , 0 ) , the generalized Laguerre polynomials L n ( ∣ k ∣ ) L_n^{(|k|)} L n ( ∣ k ∣ ) , and : : :: :: normal ordering,
ρ n ( k ) = ( − 1 ) n ( ν + 1 ) ∣ k ∣ + 1 ( a † ) k + : L n ( ∣ k ∣ ) ( n ^ ν + 1 ) e − n ^ / ( ν + 1 ) : a k − (15) \rho_n^{(k)} = \frac{(-1)^n}{(\nu+1)^{|k|+1}}(a^\dagger)^{k_+}:L_n^{(|k|)}\left(\frac{\hat{n}}{\nu+1}\right)e^{-\hat{n}/(\nu+1)}:a^{k_-} \tag*{(15)} ρ n ( k ) = ( ν + 1 ) ∣ k ∣ + 1 ( − 1 ) n ( a † ) k + : L n ( ∣ k ∣ ) ( ν + 1 n ^ ) e − n ^ / ( ν + 1 ) : a k − (15)
for n ≥ 0 n \ge0 n ≥ 0 , k ∈ Z k \in\mathbb{Z} k ∈ Z [25 ] ; it satisfies L f ρ n ( k ) = − A ( n + ∣ k ∣ / 2 ) ρ n ( k ) \mathcal{L}_f\rho_n^{(k)} = -A(n+|k|/2)\rho_n^{(k)} L f ρ n ( k ) = − A ( n + ∣ k ∣/2 ) ρ n ( k ) and tr ρ n ( k ) = δ n 0 δ k 0 \operatorname{tr}\rho_n^{(k)} = \delta_{n0}\delta_{k0} tr ρ n ( k ) = δ n 0 δ k 0 . (At ν = 0 \nu= 0 ν = 0 , : f ( n ^ ) e − n ^ : :f(\hat{n})e^{-\hat{n}}: : f ( n ^ ) e − n ^ : is evaluated with : n ^ j : = ( a † ) j ∣ 0 ⟩ ⟨ 0 ∣ a j :\hat{n}^j: = (a^\dagger)^j|0\rangle\langle0|a^j : n ^ j := ( a † ) j ∣0 ⟩ ⟨ 0∣ a j .) Writing a ( − 1 ) = a a^{(-1)} = a a ( − 1 ) = a , a ( + 1 ) = a † a^{(+1)} = a^\dagger a ( + 1 ) = a † , the field multiplications are
a ( μ ) ρ n ( k ) = ∑ δ f μ , δ L ( n , k ) ρ n + δ ( k + μ ) , a^{(\mu)}\rho_n^{(k)} = \sum_{\delta} f_{\mu,\delta}^{\mathrm{L}}(n,k)\rho_{n+\delta}^{(k+\mu)}, a ( μ ) ρ n ( k ) = δ ∑ f μ , δ L ( n , k ) ρ n + δ ( k + μ ) ,
ρ n ( k ) a ( μ ) = ∑ δ f μ , δ R ( n , k ) ρ n + δ ( k + μ ) (16) \rho_n^{(k)}a^{(\mu)} = \sum_{\delta} f_{\mu,\delta}^{\mathrm{R}}(n,k)\rho_{n+\delta}^{(k+\mu)} \tag*{(16)} ρ n ( k ) a ( μ ) = δ ∑ f μ , δ R ( n , k ) ρ n + δ ( k + μ ) (16)
δ ∈ { − 1 , 0 , 1 } \delta\in\{-1,0,1\} δ ∈ { − 1 , 0 , 1 } , ρ − 1 ( k ) = 0 \rho_{-1}^{(k)} = 0 ρ − 1 ( k ) = 0 , with the nonzero coefficients of Table VIII. Equations (F1) and (F2) and the table were checked against each other in a Fock space of dimension 70 for n ≤ 5 n \le5 n ≤ 5 , ∣ k ∣ ≤ 3 |k| \le3 ∣ k ∣ ≤ 3 , ν = 0 \nu= 0 ν = 0 and 0.3, to 2 × 10 − 15 2 \times10^{-15} 2 × 1 0 − 15 (results/check_field_coefficients.py).
b. Local structure constants. Represent an element of A \mathcal{A} A by its coordinates ( p 1 , … , p d , c + , c − ) (p_1,\ldots,p_d,c_+,c_-) ( p 1 , … , p d , c + , c − ) in { P 1 , … , P d , τ + , τ − } \{P_1,\ldots,P_d,\tau_+,\tau_-\} { P 1 , … , P d , τ + , τ − } . By the identities in the proof of Proposition 1, the five one-site maps act as
Proposition 1, the five one-site maps act as
τ + x : ( c + ′ , p ′ , c − ′ ) = ( p l , c − e u , 0 ) , x τ + : ( c + ′ , p ′ , c − ′ ) = ( p u , c − e l , 0 ) , τ − x : ( c − ′ , p ′ , c + ′ ) = ( p u , c + e l , 0 ) , x τ − : ( c − ′ , p ′ , c + ′ ) = ( p l , c + e u , 0 ) , L l o c x : p ′ = W p , c ± ′ = − ( c u l ∓ i Δ ) c ± . (17) \begin{aligned}
\tau_+x &: (c'_+,p',c'_-)=(p_l,c_-e_u,0),\\
x\tau_+ &: (c'_+,p',c'_-)=(p_u,c_-e_l,0),\\
\tau_-x &: (c'_-,p',c'_+)=(p_u,c_+e_l,0),\\
x\tau_- &: (c'_-,p',c'_+)=(p_l,c_+e_u,0),\\
\mathcal{L}_{\mathrm{loc}}x &: p'=Wp,\quad c'_{\pm}=-(c_{ul}\mp i\Delta)c_{\pm}. \tag*{(17)}
\end{aligned} τ + x x τ + τ − x x τ − L loc x : ( c + ′ , p ′ , c − ′ ) = ( p l , c − e u , 0 ) , : ( c + ′ , p ′ , c − ′ ) = ( p u , c − e l , 0 ) , : ( c − ′ , p ′ , c + ′ ) = ( p u , c + e l , 0 ) , : ( c − ′ , p ′ , c + ′ ) = ( p l , c + e u , 0 ) , : p ′ = W p , c ± ′ = − ( c u l ∓ i Δ ) c ± . (17)
with W W W of (2) and C u l C_{ul} C u l of (4) . For a basis b μ = ∑ i B i μ b_\mu=\sum_i B_{i\mu} b μ = ∑ i B i μ (coordinate vector i i i ), the structure constants of a map with coordinate matrix x x x are B − 1 x B B^{-1}xB B − 1 x B . We use the fixed basis b 0 = P g b_0=P_g b 0 = P g , b i = P i b_i=P_i b i = P i (i ≠ g i\ne g i = g ), b d + 1 = τ + b_{d+1}=\tau_+ b d + 1 = τ + , whose only traceful element has tr b 0 = 1 \operatorname{tr}b_0=1 tr b 0 = 1 , so that tr R m = δ m , N e 0 \operatorname{tr}R_m=\delta_{m,N_e0} tr R m = δ m , N e 0 . Lifting the five maps with (6) gives the D N × D N D_N\times D_N D N × D N matrices S ± L , S ± R S_\pm^{\mathrm{L}},S_\pm^{\mathrm{R}} S ± L , S ± R (left and right multiplication by S ± S_\pm S ± ) and Λ t \Lambda^t Λ t (the master Liouvillian including detuning). The charge of R m R_m R m is q = m + − m − q=m_+-m_- q = m + − m − , and in the sector K K K basis state α \alpha α carries the field grade k α = K − q α k_\alpha=K-q_\alpha k α = K − q α .
c. Blocks. With ∫ i n t ρ = ( i g / 2 ) [ a † S − + a S + , ρ ] \int_{\mathrm{int}}\rho=(ig/2)[a^\dagger S_-+aS_+,\rho] ∫ int ρ = ( i g /2 ) [ a † S − + a S + , ρ ] and the diagonal matrices f μ , δ L / R ( n ) = f μ , δ L / R ( n , k α ) f_{\mu,\delta}^{\mathrm{L/R}}(n)=f_{\mu,\delta}^{\mathrm{L/R}}(n,k_\alpha) f μ , δ L/R ( n ) = f μ , δ L/R ( n , k α ) , the coupling of X n X_n X n into the equation for X n + δ X_{n+\delta} X n + δ is
B n ( δ ) = i g 2 [ S − L f + , δ L ( n ) + S + L f − , δ L ( n ) − S − R f + , δ R ( n ) − S + R f − , δ R ( n ) ] , B_n^{(\delta)}=\frac{ig}{2}\left[S_-^{\mathrm{L}}f_{+,\delta}^{\mathrm{L}}(n)+S_+^{\mathrm{L}}f_{-,\delta}^{\mathrm{L}}(n)-S_-^{\mathrm{R}}f_{+,\delta}^{\mathrm{R}}(n)-S_+^{\mathrm{R}}f_{-,\delta}^{\mathrm{R}}(n)\right], B n ( δ ) = 2 i g [ S − L f + , δ L ( n ) + S + L f − , δ L ( n ) − S − R f + , δ R ( n ) − S + R f − , δ R ( n ) ] ,
M n = − A diag α ( n + ∣ k α ∣ 2 ) + Λ t + B n ( 0 ) , M_n=-A\operatorname{diag}_{\alpha}\left(n+\frac{|k_\alpha|}{2}\right)+\Lambda^t+B_n^{(0)}, M n = − A diag α ( n + 2 ∣ k α ∣ ) + Λ t + B n ( 0 ) ,
G n = B n + 1 ( − 1 ) , F n = B n − 1 ( + 1 ) , F 0 = 0. (18) G_n=B_{n+1}^{(-1)},\qquad F_n=B_{n-1}^{(+1)},\qquad F_0=0. \tag*{(18)} G n = B n + 1 ( − 1 ) , F n = B n − 1 ( + 1 ) , F 0 = 0. (18)
The affine coefficients of Proposition 2 are M ( 1 ) = M n + 1 − M n M^{(1)}=M_{n+1}-M_n M ( 1 ) = M n + 1 − M n , M ( 0 ) = M 0 M^{(0)}=M_0 M ( 0 ) = M 0 , F ( 1 ) = F n + 1 − F n F^{(1)}=F_{n+1}-F_n F ( 1 ) = F n + 1 − F n , and G = G n G=G_n G = G n .
d. Solution and observables. For K = 0 K=0 K = 0 : set R n max = R ∞ = − ( M ( 1 ) ) − 1 F ( 1 ) R_{n_{\max}}=R_\infty=-(M^{(1)})^{-1}F^{(1)} R n m a x = R ∞ = − ( M ( 1 ) ) − 1 F ( 1 ) , recurse R n = − ( M n + 1 + G n + 1 R n + 1 ) − 1 F n R_n=-(M_{n+1}+G_{n+1}R_{n+1})^{-1}F_n R n = − ( M n + 1 + G n + 1 R n + 1 ) − 1 F n down to R 0 R_0 R 0 , solve ( M 0 + G 0 R 0 ) X 0 = 0 (M_0+G_0R_0)X_0=0 ( M 0 + G 0 R 0 ) X 0 = 0 with the row of N e 0 N_{e0} N e 0 replaced by the normalization X 0 [ N e 0 ] = 1 X_0[N_{e0}]=1 X 0 [ N e 0 ] = 1 , and set X n + 1 = R n X n X_{n+1}=R_nX_n X n + 1 = R n X n . Observables follow from Sec. III C. For the counting statistics, the coefficient vector Y Y Y of ρ s s a ^ † j \rho_{ss}\hat{a}^{\dagger j} ρ ss a ^ † j is obtained by applying (16) twice to every component ( X n ) α (X_n)_\alpha ( X n ) α , first f − , δ 1 L ( n , k α ) f_{-,\delta_1}^{\mathrm{L}}(n,k_\alpha) f − , δ 1 L ( n , k α ) and then f + , δ 2 R ( n + δ 1 , k α − 1 ) f_{+,\delta_2}^{\mathrm{R}}(n+\delta_1,k_\alpha-1) f + , δ 2 R ( n + δ 1 , k α − 1 ) , the atomic index being unchanged; F c o u n t F_{\mathrm{count}} F count then follows from the Poisson equation of Sec. VI for the truncated block-tridiagonal matrix M \mathcal{M} M built from Eq. (18) . The emission spectrum uses the same blocks with K = − 1 K=-1 K = − 1 . As a checksum, two ladder emitters with g = 5 A g=5A g = 5 A , γ = A \gamma=A γ = A , γ p = γ r = 3 A \gamma_p=\gamma_r=3A γ p = γ r = 3 A , P = 3 A P=3A P = 3 A , Δ = ν = 0 \Delta=\nu=0 Δ = ν = 0 give ⟨ n ^ ⟩ = 0.99574218 \langle\hat{n}\rangle=0.99574218 ⟨ n ^ ⟩ = 0.99574218 , g ( 2 ) ( 0 ) = 0.97279395 g^{(2)}(0)=0.97279395 g ( 2 ) ( 0 ) = 0.97279395 , and F c o u n t = 0.73821790 F_{\mathrm{count}}=0.73821790 F count = 0.73821790 (n max = 30 n_{\max}=30 n m a x = 30 and 40 agree to 10 − 14 10^{-14} 1 0 − 14 ).
References