Introduction
In this article we discuss a recurring question in free-electron laser (FEL) theory: under what conditions can quantum signatures of the interaction be observed? The question dates back to the foundations of the subject. Madey described stimulated emission in a periodic magnetic field in terms of the exchange of radiation photons and undulator pseudophotons [1 ] . This language connects the microscopic interaction to stimulated gain, but does not by itself identify an experimentally resolvable departure from classical FEL dynamics.
Independent early quantum descriptions include the theory of Becker and Mitter [2 ] and the strong-field stimulated-process analysis of McIver and Fedorov [3 ] . The literature has developed considerably, from the early quantum analyses and the previous references collected therein [4 , 5 ] , to studies of coherent states, classical correspondence and quantum interaction [6 , 7 , 8 ] . Becker and McIver developed a fully quantized many-particle treatment of small-signal gain and photon statistics [9 ] ; their subsequent review extends the discussion to a class of free-electron devices, including optical wigglers [10 ] . Gea-Banacloche developed a many-particle quantum treatment connecting large gain, saturation and photon-number moments [11 ] ; the review by Friedman, Gover, Kurizki, Ruschin and Yariv places spontaneous and stimulated emission in the broader framework of quasifree-electron radiation [12 ] . Early work on photon statistics and squeezed radiation also illustrates the diversity of quantum-optical questions posed for the FEL [13 , 14 ] . These early analyses already connect recoil, spontaneous startup, stimulated amplification and the statistical preparation of the beam and field. Later work addresses high-gain quantum corrections [15 ] and collective amplification when discrete recoil becomes dynamically important [16 ] . Kling and collaborators identified a recoil-selective two-state regime and its quantum-optical interpretation [17 ] , then developed a momentum-space Dicke description of exponential gain [18 ] . Their subsequent long-time analysis distinguishes photon-number means from preparation-dependent fluctuations and examines experimental requirements [19 ] . These contributions concern related, but distinct, levels of description. Relevant references will be quoted in the course of the discussion. The Bambini–Renieri (BR) moving-frame formulation supplies a particularly useful bridge between the relativistic laboratory description and the effective longitudinal interaction [20 , 21 ] .
In the following, we make extensive use of the quantum analysis of the FEL developed within the Bambini-Renieri Hamiltonian framework. Early treatments reproduced the classical small-signal gain and clarified the difficulty of isolating quantum signatures. Subsequent contributions established a broad theoretical basis for a dedicated experimental programme. Raman-Nath ladder equations and recoil expansions appeared in several early treatments and became widely used; the present discussion does not assign these tools exclusively to a single contribution. The references cited here trace the development from low-gain quantum dynamics to collective high-gain models and identify the present experimental bottlenecks [2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 ] .
We distinguish single-particle quantum effects from collective quantum effects. At the single-electron level, quantized momentum exchange, the energy differences between recoil orders, and electron–photon entanglement can change transition probabilities and conditional radiation statistics. These effects can be studied before introducing beam amplification. Collective quantum effects arise when such microscopic transitions participate in the self-consistent evolution of many electrons coupled to a common radiation field: recoil can then change the gain spectrum, accessible momentum populations and correlations of the amplified field. Multiplying a single-electron emission probability by the electron number does not establish collective quantum dynamics. Conversely, bunching and collective gain already exist in classical FEL theory and are not, on their own, evidence of quantum recoil.
The observability of a quantum correction must be assessed against competing effects: electron energy spread and emittance, undulator imperfections, finite-electron shot noise, and stochastic energy changes caused by spontaneous radiation in the undulator. The importance of each effect depends on the operating regime. A quantum description of emission is necessary at the microscopic level, but observing radiation or startup does not alone demonstrate a nonclassical signature. Amplifier and multimode analyses illustrate why the optical state and its statistics must be considered together with the beam [22 , 23 ] . The appropriate comparison is therefore between quantum and classical predictions for the same measured quantity, with the same beam preparation, interaction and detection conditions.
The plan of the article follows this distinction. We first establish the Hamiltonian, its reference frame and its operator structure, explaining how the prescribed undulator enters the reduced interaction. The interaction-picture treatment then derives detuning and recoil, and identifies the prescribed-current solution as a useful reference. We develop the Raman–Nath momentum–photon ladder and its perturbative recoil corrections before discussing modulation, bunching and beam averaging. This order separates elementary quantum transitions from the collective response that may amplify or obscure them. Only afterwards do we introduce the resonator and compare its dynamics with the recoil-resolved, approximately two-state oscillator of Kling and Giese [24 ] . The comparison asks which quantum signatures survive fresh-electron pumping and cavity loss, recognizing the innovative experimental perspective of the oscillator proposal. The final appendices compare low-gain quantum dynamics with high-gain SASE and establish the shot-noise baseline for interpreting startup and radiation statistics.
The contribution of the present work is a unified, preparation-sensitive account of recoil observability. We formulate the detuning-exact local-mismatch hierarchy, validate its second-order amplitudes against direct ladder propagation, and connect it to reduced optical coherence and beam averaging. We then use these distinctions to assess resonance acceptance and photon-statistical diagnostics in an optical-undulator oscillator. These results delimit the inference available from specified observables; they do not constitute a new collective high-gain theory or a demonstration of experimental feasibility.
Relativistic starting point, gauge choice and FEL frame
The laboratory-frame starting point is the spin-independent, positive-energy minimal-coupling Hamiltonian for an electron of charge − e -e − e ,
H ^ = m 2 c 4 + c 2 [ p ^ + e A u ( r ^ ) + e A ^ r ( r ^ ) ] 2 + H ^ r a d . (1) \hat{H} = \sqrt{m^{2}c^{4} + c^{2}\left[\hat{\mathbf{p}} + e\mathbf{A}_{u}(\hat{\mathbf{r}}) + e\hat{\mathbf{A}}_{r}(\hat{\mathbf{r}})\right]^{2}} + \hat{H}_{\mathrm{rad}}. \tag*{(1)} H ^ = m 2 c 4 + c 2 [ p ^ + e A u ( r ^ ) + e A ^ r ( r ^ ) ] 2 + H ^ rad . (1)
Here m = m e m=m_{e} m = m e is the invariant electron rest mass; c c c the speed of light, A u \mathbf{A}_{u} A u the prescribed undulator vector potential and A ^ r \hat{\mathbf{A}}_{r} A ^ r the quantized radiation vector potential. Spin and pair production are outside this effective description. We use the Coulomb (radiation) gauge, ∇ ⋅ A = 0 \nabla\cdot\mathbf{A}=0 ∇ ⋅ A = 0 , with Φ = 0 \Phi=0 Φ = 0 for the transverse laboratory problem. Physical transverse photon degrees of freedom are explicit in this gauge; gauge-invariant observables do not depend on the choice. A subsequent boost and reduction must transform the fields consistently rather than retain laboratory potentials unchanged.
The reduction to one longitudinal degree of freedom must be stated carefully. In the laboratory the electron has a large longitudinal momentum, but the undulator generates an oscillating transverse mechanical momentum. For a planar undulator its characteristic magnitude is
p ⊥ m c ∼ K , K = e B u λ u 2 π m c . (2) \frac{p_{\perp}}{mc} \sim K,\qquad K = \frac{eB_{u}\lambda_{u}}{2\pi mc}. \tag*{(2)} m c p ⊥ ∼ K , K = 2 π m c e B u λ u . (2)
Here B u B_u B u and λ u \lambda_u λ u are the laboratory peak field and period. Polarization and averaging introduce the familiar convention-dependent factors. The electron–radiation coupling remains possible when the reduced dynamics is written only in terms of z z z and p z p_z p z : the transverse motion has first been generated by the undulator and eliminated in favor of an effective longitudinal interaction.
This resolves an apparent difficulty in Eq. (1) . If p ^ = p ^ z e z \hat{\mathbf{p}}=\hat{p}_z\mathbf{e}_z p ^ = p ^ z e z and both vector potentials are transverse, p ^ ⋅ A = 0 \hat{\mathbf{p}}\cdot\mathbf{A}=0 p ^ ⋅ A = 0 . The resonant interaction does not disappear, because
( A u + A ^ r ) 2 = A u 2 + A ^ r 2 + 2 A u ⋅ A ^ r . (3) \left(\mathbf{A}_u+\hat{\mathbf{A}}_r\right)^2=\mathbf{A}_u^2+\hat{\mathbf{A}}_r^2+2\mathbf{A}_u\cdot\hat{\mathbf{A}}_r. \tag*{(3)} ( A u + A ^ r ) 2 = A u 2 + A ^ r 2 + 2 A u ⋅ A ^ r . (3)
The cross term mixes undulator and radiation harmonics. In a formulation retaining the transverse mechanical momentum, the same interaction is represented through the undulator-induced p ⊥ p_\perp p ⊥ . The one-dimensional theory is thus an effective longitudinal theory obtained after the transverse dynamics has supplied the coupling. Its reduction and relation to quantum optics are discussed in [25 ] .
We now specify the moving frame explicitly. Let k r = ω r / c > 0 k_r=\omega_r/c>0 k r = ω r / c > 0 be the laboratory wave number of the forward radiation and k u = 2 π / λ u > 0 k_u=2\pi/\lambda_u>0 k u = 2 π / λ u > 0 the undulator spatial frequency. Select the static harmonic with signed wave number − k u -k_u − k u . A boost along the electron beam with velocity β B c \beta_Bc β B c and Lorentz factor γ B \gamma_B γ B gives
ω r ′ = γ B ω r ( 1 − β B ) , k r ′ = γ B k r ( 1 − β B ) , ω u ′ = γ B β B c k u , k u ′ = − γ B k u . (4) \begin{aligned}
\omega'_r&=\gamma_B\omega_r(1-\beta_B), & k'_r&=\gamma_B k_r(1-\beta_B),\\
\omega'_u&=\gamma_B\beta_B c k_u, & k'_u&=-\gamma_B k_u. \tag*{(4)}
\end{aligned} ω r ′ ω u ′ = γ B ω r ( 1 − β B ) , = γ B β B c k u , k r ′ k u ′ = γ B k r ( 1 − β B ) , = − γ B k u . (4)
The subscript B B B labels the Bambini–Renieri boost; β B \beta_B β B is its velocity divided by c c c , not the undulator magnetic field. Primes denote BR-frame quantities. Choosing β B = k r / ( k r + k u ) \beta_B=k_r/(k_r+k_u) β B = k r / ( k r + k u ) makes the two transformed frequencies equal. For a magnetic undulator, however, the transformed static harmonic remains spacelike: ( ω u ′ / c ) 2 − ( k u ′ ) 2 = − k u 2 (\omega'_u/c)^2-(k'_u)^2=-k_u^2 ( ω u ′ / c ) 2 − ( k u ′ ) 2 = − k u 2 . It is therefore not an exact free electromagnetic wave. In the ultrarelativistic boost limit it is approximately counterpropagating with the radiation, so that
ω u ′ = ω r ′ ≡ ω ′ , k u ′ ≃ − k r ′ , ∣ k u ′ ∣ ≃ k r ′ ≡ k ′ . (5) \omega'_u=\omega'_r\equiv\omega',\qquad k'_u\simeq-k'_r,\qquad|k'_u|\simeq k'_r\equiv k'. \tag*{(5)} ω u ′ = ω r ′ ≡ ω ′ , k u ′ ≃ − k r ′ , ∣ k u ′ ∣ ≃ k r ′ ≡ k ′ . (5)
The slow longitudinal electron motion in this frame motivates the BR reduction. For an optical undulator, both modes are propagating waves and equal frequencies imply equal and opposite wave vectors exactly within the two-wave model.
The positive recoil magnitude associated with creation of a radiation photon and removal of one pump quantum is
q ′ = ℏ ( k r ′ − k u ′ ) = ℏ ω ′ c ( 1 + 1 β B ) ≃ 2 ℏ k ′ . (6) q'=\hbar(k'_r-k'_u)=\frac{\hbar\omega'}{c}\left(1+\frac{1}{\beta_B}\right)\simeq2\hbar k'. \tag*{(6)} q ′ = ℏ ( k r ′ − k u ′ ) = c ℏ ω ′ ( 1 + β B 1 ) ≃ 2ℏ k ′ . (6)
The middle expression applies to the static magnetic harmonic of Eq. (4) ; the last expression is its ultrarelativistic approximation. For equal-frequency optical waves, q ′ = 2 ℏ k ′ q'=2\hbar k' q ′ = 2ℏ k ′ exactly. Thus the customary notation q = 2 ℏ k q=2\hbar k q = 2ℏ k is the same recoil relation after primes have been suppressed, not an additional momentum transfer. In the laboratory the corresponding spatial transfer is q l a b = ℏ ( k r + k u ) q_{\mathrm{lab}}=\hbar(k_r+k_u) q lab = ℏ ( k r + k u ) . These expressions refer to different frames; their transformation also involves the exchanged energy. For clarity, let δ E \delta E δ E and δ p z \delta p_z δ p z denote the energy and longitudinal momentum delivered to the fields in an exchange. Their boost from the laboratory to the BR frame is
δ E ′ = γ B ( δ E − β B c δ p z ) , δ p z ′ = γ B ( δ p z − β B δ E / c ) . \delta E'=\gamma_B(\delta E-\beta_B c\delta p_z),\qquad\delta p'_z=\gamma_B(\delta p_z-\beta_B\delta E/c). δ E ′ = γ B ( δ E − β B cδ p z ) , δ p z ′ = γ B ( δ p z − β B δ E / c ) .
For magnetic-undulator emission, δ E = ℏ ω r \delta E=\hbar\omega_r δ E = ℏ ω r and δ p z = ℏ ( k r + k u ) \delta p_z=\hbar(k_r+k_u) δ p z = ℏ ( k r + k u ) in the selected harmonic. Thus q ′ = δ p z ′ q'=\delta p'_z q ′ = δ p z ′ ; choosing β B = k r / ( k r + k u ) \beta_B=k_r/(k_r+k_u) β B = k r / ( k r + k u ) gives δ E ′ = 0 \delta E'=0 δ E ′ = 0 . This is the equal-frequency exchange in the BR description. Momentum transfer therefore transforms as part of a four-vector, not as an independent scalar.
The physical meaning of an undulator pseudophoton can now be stated. A static laboratory magnet is not a gas of real photons. In a rapidly moving frame its transverse magnetic field is accompanied by an electric field and becomes a time-dependent periodic drive. The Weizsäcker–Williams equivalent-photon viewpoint represents this drive by approximately wave-like Fourier components [26 ] . In the present application its usefulness rests on the near equality of transformed electric and magnetic field amplitudes when β B ≃ 1 \beta_B\simeq1 β B ≃ 1 , while the nonzero spacelike invariant noted above remains. It supplies a wave representation of the interaction, not a literal quantization of the magnet as freely propagating radiation.
An equivalent occupation can be estimated only after specifying the harmonic and its normalization volume. For a planar laboratory field B y = B u cos ( k u z ) B_y = B_u \cos(k_u z) B y = B u cos ( k u z ) , the transformed amplitudes obey ∣ B y ′ ∣ = γ B ∣ B y ∣ |B'_y| = \gamma_B |B_y| ∣ B y ′ ∣ = γ B ∣ B y ∣ and ∣ E x ′ ∣ = γ B β B c ∣ B y ∣ |E'_x| = \gamma_B \beta_B c |B_y| ∣ E x ′ ∣ = γ B β B c ∣ B y ∣ . The period-averaged field energy density is consequently
u ′ ‾ = ϵ 0 2 E ′ ‾ 2 + 1 2 μ 0 B ′ ‾ 2 = γ B 2 ( 1 + β B 2 ) B u 2 4 μ 0 . (7) \overline{u'} = \frac{\epsilon_0}{2}\overline{E'}^2 + \frac{1}{2\mu_0}\overline{B'}^2 = \frac{\gamma_B^2(1+\beta_B^2)B_u^2}{4\mu_0}. \tag*{(7)} u ′ = 2 ϵ 0 E ′ 2 + 2 μ 0 1 B ′ 2 = 4 μ 0 γ B 2 ( 1 + β B 2 ) B u 2 . (7)
Here ϵ 0 \epsilon_0 ϵ 0 and μ 0 \mu_0 μ 0 are the vacuum permittivity and permeability. Assigning energy ℏ ω u ′ \hbar\omega'_u ℏ ω u ′ to each equivalent quantum in a BR-frame effective mode volume V e f f ′ V'_{\mathrm{eff}} V eff ′ gives the normalization estimate
ν ‾ u e q = u ′ ‾ V e f f ′ ℏ ω u ′ = γ B ( 1 + β B 2 ) B u 2 V e f f ′ 4 μ 0 ℏ β B c k u . (8) \overline{\nu}_u^{\mathrm{eq}} = \frac{\overline{u'}V'_{\mathrm{eff}}}{\hbar\omega'_u} = \frac{\gamma_B(1+\beta_B^2)B_u^2V'_{\mathrm{eff}}}{4\mu_0\hbar\beta_B c k_u}. \tag*{(8)} ν u eq = ℏ ω u ′ u ′ V eff ′ = 4 μ 0 ℏ β B c k u γ B ( 1 + β B 2 ) B u 2 V eff ′ . (8)
This is an equivalent-mode energy estimate for the chosen planar harmonic, not a frame-independent number of photons stored in a static magnet. The volume, overlap and harmonic profile must be supplied for a numerical value. A change of mode normalization changes both the assigned occupation and the single-quantum coupling; the physical classical drive is their invariant product. For a true optical pump in a coherent state, by contrast, ν ‾ u = ∣ α u ∣ 2 \overline{\nu}_u = |\alpha_u|^2 ν u = ∣ α u ∣ 2 is the actual mean occupation of the selected propagating mode.
The review article reported in Ref. [27 ] interprets the undulator harmonic through equivalent massive pseudophotons. The invariant above corresponds formally to the momentum scale m p s = ℏ k u / c m_{\mathrm{ps}} = \hbar k_u/c m ps = ℏ k u / c , with a spacelike four-wavevector; it is not an on-shell photon of positive invariant mass. The static magnetic field is supported by its physical sources. For the Maxwell–Proca massive-vector formulation and its distinction from ordinary Maxwell electrodynamics, see Refs. [28 , 29 ] . The Proca-like analogy here organizes equivalent-quanta kinematics rather than implying a fundamental photon mass or a violation of source-inclusive Maxwell equations. In particular, the spacelike undulator harmonic must not be identified with an on-shell massive Proca particle.
For the following operator discussion we use BR coordinates and suppress primes: z = z ′ z = z' z = z ′ , p = p z ′ p = p'_z p = p z ′ and q = q ′ q = q' q = q ′ . Introducing a ^ u \hat{a}_u a ^ u for the pump or its equivalent-mode representation and a ^ \hat{a} a ^ for radiation, the resonant two-mode interaction is
H ^ i n t ( 2 ) = ℏ g u ( a ^ † a ^ u T ^ q + a ^ a ^ u T ^ q † ) , T ^ q = e − i q z / ℏ . (9) \hat{H}_{\mathrm{int}}^{(2)} = \hbar g_u\left(\hat{a}^{\dagger}\hat{a}_u\hat{T}_q + \hat{a}\hat{a}_u\hat{T}_q^{\dagger}\right), \qquad\hat{T}_q = e^{-iqz/\hbar}. \tag*{(9)} H ^ int ( 2 ) = ℏ g u ( a ^ † a ^ u T ^ q + a ^ a ^ u T ^ q † ) , T ^ q = e − i q z /ℏ . (9)
Here g u g_u g u is the effective single-pump-quantum, single-radiation-quantum coupling, with dimensions of angular frequency (s − 1 \mathrm{s}^{-1} s − 1 ). It includes polarization, harmonic and spatial overlap. Its value is fixed by the field normalization and the transverse reduction; it is not an independent universal constant. For two equal-frequency transverse plane-wave modes in the nonrelativistic BR reduction, define their vector-potential amplitudes per quantum by
A j , z p f = ℏ 2 ϵ 0 ω u ′ V j ′ , g u = e 2 ξ A u , z p f A r , z p f m ℏ = 2 π α f s ℏ c ξ m ω u ′ V u ′ V r ′ , α f s = e 2 4 π ϵ 0 ℏ c . (10) A_{j,\mathrm{zpf}} = \sqrt{\frac{\hbar}{2\epsilon_0\omega'_u V'_j}}, \qquad g_u = \frac{e^2\xi A_{u,\mathrm{zpf}}A_{r,\mathrm{zpf}}}{m\hbar} = \frac{2\pi\alpha_{\mathrm{fs}}\hbar c\xi}{m\omega'_u\sqrt{V'_uV'_r}}, \qquad\alpha_{\mathrm{fs}} = \frac{e^2}{4\pi\epsilon_0\hbar c}. \tag*{(10)} A j , zpf = 2 ϵ 0 ω u ′ V j ′ ℏ , g u = m ℏ e 2 ξ A u , zpf A r , zpf = m ω u ′ V u ′ V r ′ 2 π α fs ℏ c ξ , α fs = 4 π ϵ 0 ℏ c e 2 . (10)
Equivalently, with the reduced Compton wavelength λ ˉ C = ℏ / ( m c ) \bar{\lambda}_C = \hbar/(mc) λ ˉ C = ℏ/ ( m c ) and the common BR-frame wavelength λ ′ = 2 π c / ω ′ \lambda' = 2\pi c/\omega' λ ′ = 2 π c / ω ′ , the same coupling is
g u = α f s c λ ˉ C λ ′ ξ V u ′ V r ′ = α f s c λ C λ ′ ξ 2 π V u ′ V r ′ , λ C = h m c = 2 π λ ˉ C . (11) g_u = \frac{\alpha_{\mathrm{fs}}c\bar{\lambda}_C\lambda'\xi}{\sqrt{V'_uV'_r}} = \frac{\alpha_{\mathrm{fs}}c\lambda_C\lambda'\xi}{2\pi\sqrt{V'_uV'_r}}, \qquad\lambda_C = \frac{h}{mc} = 2\pi\bar{\lambda}_C. \tag*{(11)} g u = V u ′ V r ′ α fs c λ ˉ C λ ′ ξ = 2 π V u ′ V r ′ α fs c λ C λ ′ ξ , λ C = m c h = 2 π λ ˉ C . (11)
Both wavelengths and the volumes in this expression are dimensional SI quantities; g u g_u g u is in s − 1 \mathrm{s}^{-1} s − 1 . The factor of 2 π 2\pi 2 π depends only on whether the ordinary or reduced Compton wavelength is used.
The dimensionless overlap ξ \xi ξ includes the polarization scalar product and the selected Fourier harmonic; the phase may be absorbed in the mode operators. This expression follows from the resonant part of e 2 A u ⋅ A r / m e^2 A_u \cdot A_r/m e 2 A u ⋅ A r / m . It assumes negligible transverse mass dressing; a dressed reduction changes the effective coefficient accordingly. For a magnetic-undulator pseudomode, the wave-like normalization is only the equivalent-mode approximation described above. A classical pump replaces g u a ^ u g_u\hat{a}_u g u a ^ u by g u α u g_u\alpha_u g u α u , so the measurable reduced coupling is g = g u ∣ α u ∣ g = g_u|\alpha_u| g = g u ∣ α u ∣ , independent of the arbitrary pump-mode normalization.
The operators act on the tensor-product space
H = L 2 ( R ) e ⊗ F r ⊗ F u , [ z ^ , p ^ ] = i ℏ , [ p ^ , T ^ q ] = − q T ^ q . (12) \mathcal{H}=L^2(\mathbb{R})_{\mathrm e}\otimes\mathcal{F}_{\mathrm r}\otimes\mathcal{F}_{\mathrm u},\qquad[\hat{z},\hat{p}]=i\hbar,\qquad[\hat{p},\hat{T}_q]=-q\hat{T}_q. \tag*{(12)} H = L 2 ( R ) e ⊗ F r ⊗ F u , [ z ^ , p ^ ] = i ℏ , [ p ^ , T ^ q ] = − q T ^ q . (12)
The electron coordinate and momentum act on the first factor; the two bosonic modes act on their respective Fock spaces. The unitary operator T ^ q \hat{T}_q T ^ q translates electron momentum and commutes with the field operators. Momentum kets are generalized eigenstates; physical wave packets are superpositions of them. The elementary transition and its bosonic matrix-element factor are
∣ p , n , ν u ⟩ ⟶ ∣ p − q , n + 1 , ν u − 1 ⟩ , n + 1 ν u . (13) |p,n,\nu_u\rangle\longrightarrow|p-q,n+1,\nu_u-1\rangle,\qquad\sqrt{n+1}\sqrt{\nu_u}. \tag*{(13)} ∣ p , n , ν u ⟩ ⟶ ∣ p − q , n + 1 , ν u − 1 ⟩ , n + 1 ν u . (13)
Emission is therefore a joint transition of electron and field, not a photon creation event with an unchanged electron. The interaction possesses a dressed Schwinger s u ( 2 ) su(2) s u ( 2 ) algebra, developed in the Bloch-like treatment [30 ] and reviewed in [27 ] . Define
J ^ + = a ^ † a ^ u T ^ q , J ^ − = a ^ a ^ u † T ^ q † , J ^ z = 1 2 ( n ^ r − n ^ u ) . (14) \hat{J}_{+}=\hat{a}^{\dagger}\hat{a}_u\hat{T}_q,\qquad\hat{J}_{-}=\hat{a}\hat{a}_u^{\dagger}\hat{T}_q^{\dagger},\qquad\hat{J}_z=\frac{1}{2}(\hat{n}_r-\hat{n}_u). \tag*{(14)} J ^ + = a ^ † a ^ u T ^ q , J ^ − = a ^ a ^ u † T ^ q † , J ^ z = 2 1 ( n ^ r − n ^ u ) . (14)
Unitarity of T ^ q \hat{T}_q T ^ q and the bosonic commutators give
[ J ^ z , J ^ ± ] = ± J ^ ± , [ J ^ + , J ^ − ] = 2 J ^ z , H ^ i n t ( 2 ) = ℏ g u ( J ^ + + J ^ − ) . (15) [\hat{J}_z,\hat{J}_{\pm}]=\pm\hat{J}_{\pm},\qquad[\hat{J}_{+},\hat{J}_{-}]=2\hat{J}_z,\qquad\hat{H}_{\mathrm{int}}^{(2)}=\hbar g_u(\hat{J}_{+}+\hat{J}_{-}). \tag*{(15)} [ J ^ z , J ^ ± ] = ± J ^ ± , [ J ^ + , J ^ − ] = 2 J ^ z , H ^ int ( 2 ) = ℏ g u ( J ^ + + J ^ − ) . (15)
The total mode occupation N ^ = n ^ r + n ^ u \hat{N}=\hat{n}_r+\hat{n}_u N ^ = n ^ r + n ^ u and the combined recoil label P ^ = p ^ + q n ^ r \hat{P}=\hat{p}+q\hat{n}_r P ^ = p ^ + q n ^ r commute with this interaction. At fixed N = n 0 + ν u 0 N=n_0+\nu_{u0} N = n 0 + ν u 0 and P = p 0 + q n 0 P=p_0+qn_0 P = p 0 + q n 0 , the coupled states are
∣ p 0 − ℓ q , n 0 + ℓ , ν u 0 − ℓ ⟩ , − n 0 ≤ ℓ ≤ ν u 0 . (16) |p_0-\ell q,n_0+\ell,\nu_{u0}-\ell\rangle,\qquad-n_0\leq\ell\leq\nu_{u0}. \tag*{(16)} ∣ p 0 − ℓ q , n 0 + ℓ , ν u 0 − ℓ ⟩ , − n 0 ≤ ℓ ≤ ν u 0 . (16)
This is a finite ladder with N + 1 N+1 N + 1 states, carrying spin j = N / 2 j=N/2 j = N /2 . Electron wave packets involve a continuum of such conserved-P P P sectors. The algebra thus organizes joint recoil and mode-exchange states; it does not reduce the electron’s full Hilbert space to a single spin.
The kinetic term is essential to their dynamics. As emphasized in [27 ] , the full electron–field algebra also contains momentum translations; the full Hamiltonian is not simply a spin in a prescribed magnetic field. A Bloch-like representation alone therefore does not remove the kinetic recoil contribution. For a longitudinal dispersion E ( p ) E(p) E ( p ) ,
[ E ( p ^ ) , J ^ + ] ∣ p , n , ν u ⟩ = [ E ( p − q ) − E ( p ) ] J ^ + ∣ p , n , ν u ⟩ . (17) [E(\hat{p}),\hat{J}_{+}]|p,n,\nu_u\rangle=[E(p-q)-E(p)]\hat{J}_{+}|p,n,\nu_u\rangle. \tag*{(17)} [ E ( p ^ ) , J ^ + ] ∣ p , n , ν u ⟩ = [ E ( p − q ) − E ( p )] J ^ + ∣ p , n , ν u ⟩ . (17)
It generally does not commute with the interaction. The free two-mode field energy additionally contributes ℏ ( ω r ′ − ω u ′ ) \hbar(\omega'_r-\omega'_u) ℏ ( ω r ′ − ω u ′ ) to the energy difference of each exchange. The nonlinear dependence of E ( p ) E(p) E ( p ) on momentum generates recoil-dependent phases, so closure of the interaction operators as s u ( 2 ) su(2) s u ( 2 ) does not imply that the full Hamiltonian is linear in those generators. Deriving these phases is the purpose of the next section.
For a macroscopic prescribed pump we replace its amplitude by a classical value,
a ^ u ⟶ ν ˉ u e i ϕ u , g = g u ν ˉ u , N e x c h a n g e d ≪ ν ˉ u . (18) \hat{a}_u\longrightarrow\sqrt{\bar{\nu}_u}e^{i\phi_u},\qquad g=g_u\sqrt{\bar{\nu}_u},\qquad N_{\mathrm{exchanged}}\ll\bar{\nu}_u. \tag*{(18)} a ^ u ⟶ ν ˉ u e i ϕ u , g = g u ν ˉ u , N exchanged ≪ ν ˉ u . (18)
For a static magnet, ν ˉ u \bar{\nu}_u ν ˉ u has the equivalent-mode meaning of Eq. (8) ; only the resulting classical drive is physical. For an optical pump the replacement assumes a large coherent amplitude with negligible depletion. Its phase can be absorbed into the radiation operator. This assumption removes the finite two-mode exchange symmetry and the dynamical pump occupation. It is a controlled large-pump reduction of the interaction, not an algebra-preserving substitution. The remaining dressed radiation operators b ^ † = a ^ † T ^ q \hat{b}^{\dagger}=\hat{a}^{\dagger}\hat{T}_q b ^ † = a ^ † T ^ q and b ^ = a ^ T ^ q † \hat{b}=\hat{a}\hat{T}_q^{\dagger} b ^ = a ^ T ^ q † obey [ b ^ , b ^ † ] = 1 [\hat{b},\hat{b}^{\dagger}]=1 [ b ^ , b ^ † ] = 1 : an oscillator algebra replaces the finite s u ( 2 ) su(2) s u ( 2 ) ladder.
After the transverse reduction, resonant-harmonic selection and undepleted-pump approximation, the one-mode Hamiltonian is
H ^ e f f = E ( p ^ ) + ℏ ω a ^ † a ^ + ℏ g ( a ^ † T ^ q + a ^ T ^ q † ) . (19) \hat{H}_{\mathrm{eff}}=E(\hat{p})+\hbar\omega\hat{a}^{\dagger}\hat{a}+\hbar g\left(\hat{a}^{\dagger}\hat{T}_q+\hat{a}\hat{T}_q^{\dagger}\right). \tag*{(19)} H ^ eff = E ( p ^ ) + ℏ ω a ^ † a ^ + ℏ g ( a ^ † T ^ q + a ^ T ^ q † ) . (19)
The dimensional coupling g g g has SI units s − 1 \mathrm{s}^{-1} s − 1 . The meaning of ω \omega ω must be specified along with the frame and energy reference. In a static laboratory-undulator description it is the radiation frequency ω r \omega_r ω r , with q = q l a b q=q_{\mathrm{lab}} q = q lab . In the BR two-mode description, the free field energy can be written ℏ ω u N ^ + ℏ ( ω r ′ − ω u ′ ) n ^ r \hbar\omega_u\hat{N}+\hbar(\omega'_r-\omega'_u)\hat{n}_r ℏ ω u N ^ + ℏ ( ω r ′ − ω u ′ ) n ^ r . Removing the conserved first term, or equivalently retaining the pump phase in a rotating representation before the classical replacement, gives ω = ω r ′ − ω u ′ \omega=\omega'_r-\omega'_u ω = ω r ′ − ω u ′ . At equal-frequency BR resonance this coefficient vanishes, although the physical optical frequency ω r ′ \omega'_r ω r ′ does not. Treating ω ′ \omega' ω ′ as the remaining photon-number coefficient would incorrectly omit the pump energy from each transition.
Equation (19) is consequently a common form for explicitly specified effective descriptions. All quantities in any calculation must use the same frame and energy reference. Below the generic dispersion E ( p ) = m 2 c 4 + c 2 p 2 E(p)=\sqrt{m^2c^4+c^2p^2} E ( p ) = m 2 c 4 + c 2 p 2 displays the longitudinal recoil construction; for a slow longitudinal electron in the BR frame it reduces to m c 2 + p 2 / ( 2 m ) mc^2+p^2/(2m) m c 2 + p 2 / ( 2 m ) . If transverse averaging changes the effective dispersion, its derivatives must be evaluated from that dispersion. A laboratory relativistic factor must not be inserted independently into a BR kinetic term.
The omitted physics comprises off-resonant and counter-rotating harmonics, higher undulator harmonics, and transverse dynamics already averaged out. Nonresonant energy shifts must be included in the adopted dispersion or energy reference where relevant. These omissions define the model. Finally,
T ^ q ∣ p ⟩ = ∣ p − q ⟩ (20) \hat{T}_q|p\rangle=|p-q\rangle \tag*{(20)} T ^ q ∣ p ⟩ = ∣ p − q ⟩ (20)
identifies the reduced emission event as ∣ p , n ⟩ → ∣ p − q , n + 1 ⟩ |p,n\rangle\rightarrow|p-q,n+1\rangle ∣ p , n ⟩ → ∣ p − q , n + 1 ⟩ . Its momentum translation and kinetic-energy change inherit their meaning from the full undulator–radiation exchange.
Interaction picture, relativistic recoil and the prescribed-current limit
The interaction picture separates the free electron and radiation evolution from their coupling. Starting from Eq. (19), write
H ^ = H ^ 0 + H ^ i n t , H ^ 0 = E ( p ^ ) + H ^ r , H ^ r = ℏ ω a ^ † a ^ , (21) \hat{H}=\hat{H}_0+\hat{H}_{\mathrm{int}},\qquad\hat{H}_0=E(\hat{p})+\hat{H}_r,\qquad\hat{H}_r=\hbar\omega\hat{a}^{\dagger}\hat{a}, \tag*{(21)} H ^ = H ^ 0 + H ^ int , H ^ 0 = E ( p ^ ) + H ^ r , H ^ r = ℏ ω a ^ † a ^ , (21)
with
E ( p ) = m 2 c 4 + c 2 p 2 , H ^ i n t = ℏ g ( a ^ † T ^ q + a ^ T ^ q † ) , T ^ q = e − i q z ^ / ℏ . (22) E(p)=\sqrt{m^2c^4+c^2p^2},\qquad\hat{H}_{\mathrm{int}}=\hbar g(\hat{a}^{\dagger}\hat{T}_q+\hat{a}\hat{T}_q^{\dagger}),\qquad\hat{T}_q=e^{-iq\hat{z}/\hbar}. \tag*{(22)} E ( p ) = m 2 c 4 + c 2 p 2 , H ^ int = ℏ g ( a ^ † T ^ q + a ^ T ^ q † ) , T ^ q = e − i q z ^ /ℏ . (22)
Here E ( p ) E(p) E ( p ) denotes the longitudinal dispersion in the adopted frame and reduction. The displayed square root is the relativistic one-dimensional reference dispersion, not the slow-electron BR kinetic approximation. To reconcile this expression with the results obtained in Ref. [27 ] , we use E B R ( p ) = m c 2 + p 2 / ( 2 m ) E_{\mathrm{BR}}(p)=mc^2+p^2/(2m) E BR ( p ) = m c 2 + p 2 / ( 2 m ) . The additive rest energy is the same before and after an exchange and therefore cancels when their energy difference is formed. The subsequent ordered-operator derivation applies to either choice. The wavevector transfer q = q ′ q=q' q = q ′ and time are BR quantities when using the BR Hamiltonian; laboratory quantities are used only in a separately stated laboratory evaluation.
Following the standard interaction-picture construction [31 ] , for any Schrödinger-picture operator,
O ^ I ( t ) = e i H ^ 0 t / ℏ O ^ e − i H ^ 0 t / ℏ . (23) \hat{O}_{I}(t)=e^{i\hat{H}_0t/\hbar}\hat{O}e^{-i\hat{H}_0t/\hbar}. \tag*{(23)} O ^ I ( t ) = e i H ^ 0 t /ℏ O ^ e − i H ^ 0 t /ℏ . (23)
The electron and field operators act in different Hilbert spaces, so
[ E ( p ^ ) , H ^ r ] = 0 , e − i H ^ 0 t / ℏ = e − i E ( p ^ ) t / ℏ e − i H ^ r t / ℏ . (24) [E(\hat{p}),\hat{H}_r]=0,\qquad e^{-i\hat{H}_0t/\hbar}=e^{-iE(\hat{p})t/\hbar}e^{-i\hat{H}_rt/\hbar}. \tag*{(24)} [ E ( p ^ ) , H ^ r ] = 0 , e − i H ^ 0 t /ℏ = e − i E ( p ^ ) t /ℏ e − i H ^ r t /ℏ . (24)
Consequently,
a ^ I ( t ) = a ^ e − i ω t , a ^ I † ( t ) = a ^ † e i ω t . (25) \hat{a}_{I}(t)=\hat{a}e^{-i\omega t},\qquad\hat{a}^{\dagger}_{I}(t)=\hat{a}^{\dagger}e^{i\omega t}. \tag*{(25)} a ^ I ( t ) = a ^ e − iω t , a ^ I † ( t ) = a ^ † e iω t . (25)
The electron translation requires more care. Its exact interaction-picture form is
T ^ q , I ( t ) = e i E ( p ^ ) t / ℏ e − i q z ^ / ℏ e − i E ( p ^ ) t / ℏ . (26) \hat{T}_{q,I}(t)=e^{iE(\hat{p})t/\hbar}e^{-iq\hat{z}/\hbar}e^{-iE(\hat{p})t/\hbar}. \tag*{(26)} T ^ q , I ( t ) = e i E ( p ^ ) t /ℏ e − i q z ^ /ℏ e − i E ( p ^ ) t /ℏ . (26)
Acting on a momentum eigenstate gives successively
T ^ q , I ( t ) ∣ p ⟩ = e i E ( p ^ ) t / ℏ e − i q z ^ / ℏ e − i E ( p ) t / ℏ ∣ p ⟩ = e i [ E ( p − q ) − E ( p ) ] t / ℏ ∣ p − q ⟩ . (27) \begin{aligned}
\hat{T}_{q,I}(t)|p\rangle&= e^{iE(\hat{p})t/\hbar}e^{-iq\hat{z}/\hbar}e^{-iE(p)t/\hbar}|p\rangle\\
&= e^{i[E(p-q)-E(p)]t/\hbar}|p-q\rangle. \tag*{(27)}
\end{aligned} T ^ q , I ( t ) ∣ p ⟩ = e i E ( p ^ ) t /ℏ e − i q z ^ /ℏ e − i E ( p ) t /ℏ ∣ p ⟩ = e i [ E ( p − q ) − E ( p )] t /ℏ ∣ p − q ⟩ . (27)
Thus the free transformation itself supplies the energy difference. Including the photon creation phase leads to
Δ E ( p , q ) = E ( p − q ) − E ( p ) + ℏ ω , Δ ( p , q ) = Δ E ( p , q ) ℏ . (28) \Delta E(p,q)=E(p-q)-E(p)+\hbar\omega,\qquad\Delta(p,q)=\frac{\Delta E(p,q)}{\hbar}. \tag*{(28)} Δ E ( p , q ) = E ( p − q ) − E ( p ) + ℏ ω , Δ ( p , q ) = ℏ Δ E ( p , q ) . (28)
Here Δ E \Delta E Δ E is an energy mismatch, measured in joules, while Δ \Delta Δ is an angular-frequency mismatch, measured in s − 1 \mathrm{s}^{-1} s − 1 . This convention will also be used in spectral kernels and dephasing estimates. Detuning is not inserted as an unrelated phenomenological term: it originates in the free evolution. Before any source approximation, the exact interaction-picture Hamiltonian of this reduced model is
H ^ I ( t ) = ℏ g a † e i ω t e i E ( p ^ ) t / ℏ e − i q z ^ / ℏ e − i E ( p ^ ) t / ℏ + ℏ g a e − i ω t e i E ( p ^ ) t / ℏ e i q z ^ / ℏ e − i E ( p ^ ) t / ℏ . (29) \begin{aligned}
\hat{H}_{I}(t)={}&\hbar g a^\dagger e^{i\omega t}e^{iE(\hat{p})t/\hbar}e^{-iq\hat{z}/\hbar}e^{-iE(\hat{p})t/\hbar}\\
&+\hbar g a e^{-i\omega t}e^{iE(\hat{p})t/\hbar}e^{iq\hat{z}/\hbar}e^{-iE(\hat{p})t/\hbar}. \tag*{(29)}
\end{aligned} H ^ I ( t ) = ℏ g a † e iω t e i E ( p ^ ) t /ℏ e − i q z ^ /ℏ e − i E ( p ^ ) t /ℏ + ℏ g a e − iω t e i E ( p ^ ) t /ℏ e i q z ^ /ℏ e − i E ( p ^ ) t /ℏ . (29)
Keeping this ordering avoids treating a quantum electron phase prematurely as a classical trajectory. Under the free electron Hamiltonian,
d z ^ I d t = i ℏ [ E ( p ^ ) , z ^ I ] = v ^ ( p ^ ) , v ^ ( p ^ ) = c 2 p ^ E ( p ^ ) , z ^ I ( t ) = z ^ + v ^ ( p ^ ) t . (30) \frac{d\hat{z}_{I}}{dt}=\frac{i}{\hbar}[E(\hat{p}),\hat{z}_{I}]=\hat{v}(\hat{p}),\qquad\hat{v}(\hat{p})=\frac{c^{2}\hat{p}}{E(\hat{p})},\qquad\hat{z}_{I}(t)=\hat{z}+\hat{v}(\hat{p})t. \tag*{(30)} d t d z ^ I = ℏ i [ E ( p ^ ) , z ^ I ] = v ^ ( p ^ ) , v ^ ( p ^ ) = E ( p ^ ) c 2 p ^ , z ^ I ( t ) = z ^ + v ^ ( p ^ ) t . (30)
However,
[ z ^ , v ^ ( p ^ ) ] = i ℏ ∂ v ∂ p ≠ 0. (31) [\hat{z},\hat{v}(\hat{p})]=i\hbar\frac{\partial v}{\partial p}\ne0. \tag*{(31)} [ z ^ , v ^ ( p ^ )] = i ℏ ∂ p ∂ v = 0. (31)
The exponential of z ^ I ( t ) \hat{z}_{I}(t) z ^ I ( t ) therefore cannot be factorized as if position and velocity were commuting numbers. The momentum-state calculation in Eq. (27) resolves this ordering issue directly. The recoil content follows from the energy expansion
E ( p − q ) = E ( p ) − q E ′ ( p ) + q 2 2 E ′ ′ ( p ) − q 3 6 E ′ ′ ′ ( p ) + ⋯ . (32) E(p-q)=E(p)-qE'(p)+\frac{q^{2}}{2}E''(p)-\frac{q^{3}}{6}E'''(p)+\cdots. \tag*{(32)} E ( p − q ) = E ( p ) − q E ′ ( p ) + 2 q 2 E ′′ ( p ) − 6 q 3 E ′′′ ( p ) + ⋯ . (32)
For the relativistic dispersion,
E ′ ( p ) = v , E ′ ′ ( p ) = 1 m γ 3 , γ = E ( p ) m c 2 . (33) E'(p)=v,\qquad E''(p)=\frac{1}{m\gamma^{3}},\qquad\gamma=\frac{E(p)}{mc^{2}}. \tag*{(33)} E ′ ( p ) = v , E ′′ ( p ) = m γ 3 1 , γ = m c 2 E ( p ) . (33)
Thus
Δ ( p , q ) = ω − q v ℏ + q 2 2 ℏ m γ 3 + O ( q 3 / ℏ ) . (34) \Delta(p,q)=\omega-\frac{qv}{\hbar}+\frac{q^{2}}{2\hbar m\gamma^{3}}+O(q^{3}/\hbar). \tag*{(34)} Δ ( p , q ) = ω − ℏ q v + 2ℏ m γ 3 q 2 + O ( q 3 /ℏ ) . (34)
The first two terms give ordinary resonance mismatch; the displayed quadratic contribution is the leading longitudinal recoil correction. The factor m γ 3 m\gamma^{3} m γ 3 is the longitudinal effective mass for this specified dispersion. For ∣ p ∣ ≪ m c |p|\ll mc ∣ p ∣ ≪ m c , expand the reference dispersion as E ( p ) = m c 2 + p 2 / ( 2 m ) − p 4 / ( 8 m 3 c 2 ) + ⋯ E(p)=mc^{2}+p^{2}/(2m)-p^{4}/(8m^{3}c^{2})+\cdots E ( p ) = m c 2 + p 2 / ( 2 m ) − p 4 / ( 8 m 3 c 2 ) + ⋯ . Differentiating twice gives E ′ ′ ( p ) = 1 / m − 3 p 2 / ( 2 m 3 c 2 ) + ⋯ E''(p)=1/m-3p^{2}/(2m^{3}c^{2})+\cdots E ′′ ( p ) = 1/ m − 3 p 2 / ( 2 m 3 c 2 ) + ⋯ . Thus E B R ′ ′ = 1 / m E''_{\mathrm{BR}}=1/m E BR ′′ = 1/ m is the leading nonrelativistic term, with relative corrections of order p 2 / ( m 2 c 2 ) p^{2}/(m^{2}c^{2}) p 2 / ( m 2 c 2 ) , not an exact identity for arbitrary momentum. If transverse quiver has been absorbed in an effective rest mass, that consistently defined mass replaces m m m throughout the reduced dispersion; no laboratory γ 3 \gamma^{3} γ 3 is inserted independently. The relativistic curvature is frame-dependent, not a scalar. In one longitudinal dimension, p ′ = γ B ( p − β B E / c ) p' = \gamma_B(p-\beta_B E/c) p ′ = γ B ( p − β B E / c ) , E ′ = γ B ( E − β B c p ) E' = \gamma_B(E-\beta_B cp) E ′ = γ B ( E − β B c p ) and v ′ = ( v − β B c ) / ( 1 − β B v / c ) v' = (v-\beta_B c)/(1-\beta_B v/c) v ′ = ( v − β B c ) / ( 1 − β B v / c ) . Differentiating along the electron dispersion gives
d v ′ d p ′ = d v / d p γ B 3 ( 1 − β B v / c ) 3 = 1 m [ γ B γ ( 1 − β B v / c ) ] 3 = 1 m γ ′ 3 . \frac{dv'}{dp'} = \frac{dv/dp}{\gamma_B^3(1-\beta_B v/c)^3} = \frac{1}{m[\gamma_B\gamma(1-\beta_B v/c)]^3} = \frac{1}{m\gamma'^3}. d p ′ d v ′ = γ B 3 ( 1 − β B v / c ) 3 d v / d p = m [ γ B γ ( 1 − β B v / c ) ] 3 1 = m γ ′3 1 .
For an electron slow in that frame, γ ′ ≃ 1 \gamma' \simeq1 γ ′ ≃ 1 and the curvature becomes 1 / m 1/m 1/ m to the same nonrelativistic accuracy. This argument concerns the one-dimensional reference dispersion; transverse reduction must still be treated consistently.
For a helical undulator the transverse quiver magnitude is constant and can be incorporated in a consistent effective-mass reduction; its magnitude must still justify any nonrelativistic expansion. For a planar undulator the transverse kinetic contribution oscillates, so a boost alone does not give the same strictly stationary one-dimensional nonrelativistic problem: transverse averaging and harmonic selection are additional approximations.
After ℓ \ell ℓ recoil events the electron momentum is p 0 − ℓ q p_0-\ell q p 0 − ℓ q , and the corresponding diagonal energy difference is
Δ E ℓ = E ( p 0 − ℓ q ) − E ( p 0 ) + ℓ ℏ ω = ℓ ( ℏ ω − q v 0 ) + ℓ 2 q 2 2 m γ 0 3 + ⋯ . (35) \Delta E_\ell= E(p_0-\ell q)-E(p_0)+\ell\hbar\omega= \ell(\hbar\omega-qv_0)+\frac{\ell^2q^2}{2m\gamma_0^3}+\cdots. \tag*{(35)} Δ E ℓ = E ( p 0 − ℓ q ) − E ( p 0 ) + ℓ ℏ ω = ℓ ( ℏ ω − q v 0 ) + 2 m γ 0 3 ℓ 2 q 2 + ⋯ . (35)
As shown below, normalization gives a term linear in the exchanged-photon index and a quadratic recoil curvature. Their distinct roles will be analyzed through neighboring-state mismatches in the Raman–Nath equation.
The limit introduced next changes the dynamical problem: it produces radiation from an external source, not self-consistent FEL amplification. For uniform prescribed motion the emission and absorption kernels have the same mismatch, hence their stimulated difference is zero. In the early literature on quantum FEL dynamics [27 ] , the corresponding recoil-free seeded result is called gainless. This refers to net stimulated amplification, not to the absence of photons emitted into an empty mode.
Only after this derivation do we introduce a prescribed current by replacing z ^ I ( t ) \hat{z}_I(t) z ^ I ( t ) with an externally specified classical trajectory z c ( t ) z_c(t) z c ( t ) . This is a physical approximation, not a change of representation. The electron no longer responds dynamically to the field: recoil back-action, electron–field entanglement and field-induced trajectory modification are removed. The Hamiltonian becomes
H ^ s r c ( I ) ( t ) = ℏ g [ a ^ † e i ω t − i q z c ( t ) / ℏ + a ^ e − i ω t + i q z c ( t ) / ℏ ] . (36) \hat{H}^{(I)}_{\mathrm{src}}(t)=\hbar g\left[\hat{a}^{\dagger}e^{i\omega t-iqz_c(t)/\hbar}+\hat{a}e^{-i\omega t+iqz_c(t)/\hbar}\right]. \tag*{(36)} H ^ src ( I ) ( t ) = ℏ g [ a ^ † e iω t − i q z c ( t ) /ℏ + a ^ e − iω t + i q z c ( t ) /ℏ ] . (36)
Define the complex driving coefficient s ( t ) s(t) s ( t ) by
s ( t ) = − i g e i ω t − i q z c ( t ) / ℏ , H ^ s r c ( I ) = i ℏ [ s ( t ) a ^ † − s ∗ ( t ) a ^ ] . (37) s(t)=-ige^{i\omega t-iqz_c(t)/\hbar}, \qquad\hat{H}^{(I)}_{\mathrm{src}}=i\hbar[s(t)\hat{a}^{\dagger}-s^*(t)\hat{a}]. \tag*{(37)} s ( t ) = − i g e iω t − i q z c ( t ) /ℏ , H ^ src ( I ) = i ℏ [ s ( t ) a ^ † − s ∗ ( t ) a ^ ] . (37)
The evolution is a displacement with a scalar ordering phase,
U ^ I ( t ) = e i Φ ( t ) D ^ [ β ( t ) ] , β ( t ) = ∫ 0 t s ( t ′ ) d t ′ , D ^ ( β ) = e β a ^ † − β ∗ a ^ . (38) \hat{U}_I(t)=e^{i\Phi(t)}\hat{D}[\beta(t)], \qquad\beta(t)=\int_0^t s(t')\,dt', \qquad\hat{D}(\beta)=e^{\beta\hat{a}^{\dagger}-\beta^*\hat{a}}. \tag*{(38)} U ^ I ( t ) = e i Φ ( t ) D ^ [ β ( t )] , β ( t ) = ∫ 0 t s ( t ′ ) d t ′ , D ^ ( β ) = e β a ^ † − β ∗ a ^ . (38)
An initially empty mode therefore becomes a coherent state and has Poisson probabilities,
∣ 0 ⟩ ⟶ e i Φ ∣ β ⟩ , P n = e − μ μ n n ! , μ = ∣ β ∣ 2 . (39) |0\rangle\longrightarrow e^{i\Phi}|\beta\rangle, \qquad P_n=e^{-\mu}\frac{\mu^n}{n!}, \qquad\mu=|\beta|^2. \tag*{(39)} ∣0 ⟩ ⟶ e i Φ ∣ β ⟩ , P n = e − μ n ! μ n , μ = ∣ β ∣ 2 . (39)
This is the prescribed-current reference problem. Setting the operator commutator in Eq. (31) to zero removes its quantum ordering correction, but the approximation also removes the electron’s dynamical response. It is therefore not a general quantum-to-classical limit. A classical electron can recoil and exchange energy with a field when its trajectory is solved self-consistently. In fact, as shown in Ref. [27 ] , Eqs. (7.61)–(7.66), a quantized field driven by such a semiclassical trajectory can retain coherent-state statistics while reproducing classical gain. Coherence preservation in that approximation must not be confused with the gainless straight-trajectory limit. It is not the self-consistent classical FEL, whose electron phases and momenta remain dynamical and can develop modulation, bunching and collective gain.
Restoring the quantum electron replaces the specified phase by momentum-dependent energy differences. Successive recoil states acquire different mismatches and the exact displacement structure is generally lost. The Raman–Nath equations used below are coupled momentum-sideband equations: they describe how a periodic electron–field interaction transfers probability between recoil orders, with the detuning and kinetic recoil phases retained. The terminology recalls diffraction by a periodic grating. Here it does not mean that the kinetic recoil term has already been discarded; that would be an additional short-interaction or recoilless approximation.
Vacuum fluctuations and spontaneous emission
The prescribed-current solution has a useful connection with the coherent radiation structure of the Bloch–Nordsieck soft-photon description [32 ] . A prescribed charge current couples linearly to the creation and annihilation operators; an initially empty mode is displaced into a coherent state. This supplies a transparent reference problem with a classical external source and a quantum radiation field.
The Bloch–Nordsieck cancellation statement is more specific: infrared-finite inclusive probabilities require the appropriate combination of real and virtual soft contributions. A single-mode FEL displacement calculation illustrates the coherent-source structure, but is not a derivation of that cancellation theorem. Nor does it justify discarding recoil when recoil is the observable of interest. The interaction-picture and phase-space organization discussed later offers a complementary perspective [33 ] .
For a dynamical electron the elementary emission and absorption matrix elements have the bosonic factors
∣ M e m ∣ 2 ∝ ( n + 1 ) ∣ ⟨ p − q ∣ T ^ q ∣ p ⟩ ∣ 2 , ∣ M a b s ∣ 2 ∝ n . (40) |M_{\mathrm{em}}|^2 \propto(n+1)\left|\langle p-q|\hat{T}_q|p\rangle\right|^2,\qquad|M_{\mathrm{abs}}|^2 \propto n. \tag*{(40)} ∣ M em ∣ 2 ∝ ( n + 1 ) ⟨ p − q ∣ T ^ q ∣ p ⟩ 2 , ∣ M abs ∣ 2 ∝ n . (40)
The extra unity follows from a ^ a ^ † = a ^ † a ^ + 1 \hat{a}\hat{a}^{\dagger}=\hat{a}^{\dagger}\hat{a}+1 a ^ a ^ † = a ^ † a ^ + 1 . It is the vacuum contribution to emission into an allowed empty mode, not a pre-existing real photon in the vacuum. It arises from the nonzero vacuum matrix element of the creation operator; it is the transition probability that is nonzero. Electron kinematics, mode structure, undulator coupling and interaction duration still determine the probability.
Using the frequency mismatch Δ \Delta Δ defined in Eq. (28) , one may write schematic emission and absorption kernels as
W e m = Γ + ( p , Δ ) ( n + 1 ) , W a b s = Γ − ( p , Δ ) n . (41) W_{\mathrm{em}}=\Gamma_{+}(p,\Delta)(n+1),\qquad W_{\mathrm{abs}}=\Gamma_{-}(p,\Delta)n. \tag*{(41)} W em = Γ + ( p , Δ ) ( n + 1 ) , W abs = Γ − ( p , Δ ) n . (41)
For constant coupling over a physical duration T T T , define the per-bosonic-factor probabilities
Γ + ( p , T ) = g 2 T 2 sinc 2 [ Δ + ( p ) T / 2 ] , Γ − ( p , T ) = g 2 T 2 sinc 2 [ Δ − ( p ) T / 2 ] , (42) \Gamma_{+}(p,T)=g^2T^2\operatorname{sinc}^2\left[\Delta_{+}(p)T/2\right],\qquad\Gamma_{-}(p,T)=g^2T^2\operatorname{sinc}^2\left[\Delta_{-}(p)T/2\right], \tag*{(42)} Γ + ( p , T ) = g 2 T 2 sinc 2 [ Δ + ( p ) T /2 ] , Γ − ( p , T ) = g 2 T 2 sinc 2 [ Δ − ( p ) T /2 ] , (42)
where Δ + = ω + [ E ( p − q ) − E ( p ) ] / ℏ \Delta_{+}=\omega+[E(p-q)-E(p)]/\hbar Δ + = ω + [ E ( p − q ) − E ( p )] /ℏ and Δ − = [ E ( p + q ) − E ( p ) ] / ℏ − ω \Delta_{-}=[E(p+q)-E(p)]/\hbar-\omega Δ − = [ E ( p + q ) − E ( p )] /ℏ − ω . The even sinc makes either overall mismatch sign equivalent. Here W e m , a b s W_{\mathrm{em,abs}} W em , abs in Eq. (41) denote probabilities, not rates; these kernels retain p p p , q q q and T T T , with mismatch dependence explicit. A shaped interaction replaces g T sinc ( Δ T / 2 ) gT\operatorname{sinc}(\Delta T/2) g T sinc ( Δ T /2 ) by the appropriate time integral.
A finite-time probability is not a constant rate. Dividing by T T T gives a duration-dependent average rate, not a memoryless (Markov) population-evolution law. A time-independent rate requires an additional continuum or reservoir limit and a coarse-graining interval longer than correlation times but shorter than the population-evolution time. A selected coherent mode can instead display reversible exchange and must retain its amplitudes. Explicit dimensionless finite-interaction kernels are derived in the next section. This keeps the spontaneous factor separate from the dynamical line shape. Further comments on accumulated recoil and the prescribed-current approximation are given in the concluding section.
Photon-number statistics and optical coherence must also be distinguished. An electron in a sharp momentum state can become entangled with the radiation through distinct recoil orders. Tracing over distinguishable electron states may destroy optical off-diagonal coherences, even when the resulting photon probabilities happen to be Poissonian. The coherent state in Eq. (39) relies on the prescribed-current approximation; it is not a conclusion about every recoilless electron–field preparation.
Raman–Nath equation and local recoil expansion
We first solve an initial-value problem for a specified input state: an electron with momentum p 0 p_0 p 0 and a radiation mode containing exactly n n n photons. According to the kinematics described above, the joint electron–field state can be expanded as
∣ Ψ ( t ) ⟩ = ∑ ℓ ≥ − n C ℓ ( t ) ∣ p 0 − ℓ q , n + ℓ ⟩ . (43) |\Psi(t)\rangle=\sum_{\ell\geq-n}C_{\ell}(t)|p_0-\ell q,n+\ell\rangle. \tag*{(43)} ∣Ψ ( t )⟩ = ℓ ≥ − n ∑ C ℓ ( t ) ∣ p 0 − ℓ q , n + ℓ ⟩ . (43)
The initial condition for this preparation is
C ℓ ( 0 ) = δ ℓ 0 , ∣ Ψ ( 0 ) ⟩ = ∣ p 0 ⟩ ⊗ ∣ n ⟩ . (44) C_{\ell}(0)=\delta_{\ell0},\qquad|\Psi(0)\rangle=|p_0\rangle\otimes|n\rangle. \tag*{(44)} C ℓ ( 0 ) = δ ℓ 0 , ∣Ψ ( 0 )⟩ = ∣ p 0 ⟩ ⊗ ∣ n ⟩ . (44)
It fixes an initially unexchanged electron–photon state. A sharp momentum ket is an idealized, delocalized state, understood through box normalization or as the narrow-packet limit; it is not a general incoming electron state. Experimentally that state results from cathode emission, extraction, acceleration, bending and transport magnets, apertures and fluctuations in those stages. “Preparation” below is shorthand for the resulting measured density operator at the interaction entrance, not an assumption that an ideal state can simply be prescribed. Likewise, a photon number state is a legitimate conditional quantum state, but for a phase-defined injected laser a coherent state is usually the more appropriate input. The vacuum used below is both the number state ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ and the coherent state with zero amplitude. Choosing it removes the optical preparation ambiguity, while leaving the electron preparation to be specified separately.
A set of conditional ladder solutions can also construct a general product input. Let ϕ ( p ) \phi(p) ϕ ( p ) be a normalized electron momentum amplitude and c n c_n c n the optical number-state amplitudes. Then
∣ Ψ ( t ) ⟩ = ∑ n = 0 ∞ c n ∫ d p ϕ ( p ) ∑ ℓ ≥ − n A ℓ ( n ) ( p , t ) ∣ p − ℓ q , n + ℓ ⟩ , \lvert\Psi(t)\rangle= \sum_{n=0}^{\infty} c_n \int dp\,\phi(p) \sum_{\ell\geq-n} A_{\ell}^{(n)}(p,t)\lvert p-\ell q,n+\ell\rangle, ∣ Ψ ( t )⟩ = n = 0 ∑ ∞ c n ∫ d p ϕ ( p ) ℓ ≥ − n ∑ A ℓ ( n ) ( p , t ) ∣ p − ℓ q , n + ℓ ⟩ ,
A ℓ ( n ) ( p , t ) = e − i [ E ( p ) + n ℏ ω ] t / ℏ C ℓ ( n ) ( p , t / T 0 ) , C ℓ ( n ) ( p , 0 ) = δ ℓ 0 . (45) A_{\ell}^{(n)}(p,t)=e^{-i[E(p)+n\hbar\omega]t/\hbar}C_{\ell}^{(n)}(p,t/T_0),\qquad C_{\ell}^{(n)}(p,0)=\delta_{\ell0}. \tag*{(45)} A ℓ ( n ) ( p , t ) = e − i [ E ( p ) + n ℏ ω ] t /ℏ C ℓ ( n ) ( p , t / T 0 ) , C ℓ ( n ) ( p , 0 ) = δ ℓ 0 . (45)
Here ∫ ∣ ϕ ( p ) ∣ 2 d p = 1 \int\lvert\phi(p)\rvert^2dp=1 ∫ ∣ ϕ ( p ) ∣ 2 d p = 1 and ∑ n ∣ c n ∣ 2 = 1 \sum_n\lvert c_n\rvert^2=1 ∑ n ∣ c n ∣ 2 = 1 . For a coherent input ∣ α 0 ⟩ \lvert\alpha_0\rangle ∣ α 0 ⟩ , c n = e − ∣ α 0 ∣ 2 / 2 α 0 n / n ! c_n=e^{-\lvert\alpha_0\rvert^2/2}\alpha_0^n/\sqrt{n!} c n = e − ∣ α 0 ∣ 2 /2 α 0 n / n ! . Each conditional sector obeys its own ladder equation, with the detuning and recoil curvature evaluated at its initial momentum. The factor multiplying C C C restores the common free phase removed within that sector; across a wave packet this phase depends on p p p and cannot be discarded. Coherent optical sectors must be superposed at the amplitude level, including their relative phases, before tracing over the electron. Averaging number-state probabilities alone would instead describe a phase-randomized optical preparation. Mixed electron or optical inputs are treated by evolving their density matrices linearly. The consequences of a finite electron packet are developed in Section VI before introducing its Wigner representation.
The diagonal mismatch is already given by Eq. (35) ; it need not be expanded a second time. To fix every normalization, choose a positive reference interaction time T 0 T_0 T 0 in seconds. It can be chosen as the transit duration through the undulator. For an interaction length L u = N u λ u L_u=N_u\lambda_u L u = N u λ u in the laboratory, a central electron with velocity v 0 v_0 v 0 has T t r , l a b = L u / v 0 T_{\mathrm{tr,lab}}=L_u/v_0 T tr , lab = L u / v 0 . The duration along the same orbit in the BR frame is T t r , B R = γ B ( T t r , l a b − β B L u / c ) T_{\mathrm{tr,BR}}=\gamma_B(T_{\mathrm{tr,lab}}-\beta_B L_u/c) T tr , BR = γ B ( T tr , lab − β B L u / c ) . Because the undulator boundaries move in the BR frame, the transit is T t r , B R = ( L u , B R / v 0 ′ + β B c ) T_{\mathrm{tr,BR}}=(L_{u,\mathrm{BR}}/v'_0+\beta_B c) T tr , BR = ( L u , BR / v 0 ′ + β B c ) , where v 0 ′ = ( v 0 − β B c ) / ( 1 − β B v 0 / c ) v'_0=(v_0-\beta_B c)/(1-\beta_Bv_0/c) v 0 ′ = ( v 0 − β B c ) / ( 1 − β B v 0 / c ) . Writing L u , B R / v 0 , B R L_{u,\mathrm{BR}}/v_{0,\mathrm{BR}} L u , BR / v 0 , BR is correct only if the denominator denotes the electron velocity relative to the moving undulator boundaries, not its BR velocity alone. This reproduces the Lorentz-transformed duration above. If T 0 = T t r T_0=T_{\mathrm{tr}} T 0 = T tr in the adopted frame, the exit occurs at τ = 1 \tau=1 τ = 1 . Alternatively T 0 T_0 T 0 is a free positive normalization scale, in which case the physical exit is at τ = T t r / T 0 \tau=T_{\mathrm{tr}}/T_0 τ = T tr / T 0 ; changing T 0 T_0 T 0 changes the dimensionless parameters but not the dynamics at fixed physical time. A packet narrow enough that its transit-time variation is negligible can use the central-orbit duration. Define
τ = t T 0 , g = g T 0 , W = ( q v 0 ℏ − ω ) T 0 , ε = q 2 T 0 2 ℏ m γ 0 3 . (46) \tau=\frac{t}{T_0},\qquad g=gT_0,\qquad W=\left(\frac{qv_0}{\hbar}-\omega\right)T_0,\qquad\varepsilon=\frac{q^2T_0}{2\hbar m\gamma_0^3}. \tag*{(46)} τ = T 0 t , g = g T 0 , W = ( ℏ q v 0 − ω ) T 0 , ε = 2ℏ m γ 0 3 q 2 T 0 . (46)
For q = ℏ Q q=\hbar Q q = ℏ Q with a geometry-defined spatial wave number Q Q Q , W = ( Q v 0 − ω ) T 0 W=(Qv_0-\omega)T_0 W = ( Q v 0 − ω ) T 0 is independent of Planck’s constant. By contrast ε = ℏ Q 2 E ′ ′ T 0 / 2 \varepsilon=\hbar Q^2E''T_0/2 ε = ℏ Q 2 E ′′ T 0 /2 contains the recoil scale. Thus τ \tau τ , g g g , W W W and ε \varepsilon ε are dimensionless; g g g retains its earlier dimensional meaning. For a different effective dispersion, the last definition is ε = q 2 E ′ ′ ( p 0 ) T 0 / ( 2 ℏ ) \varepsilon=q^2E''(p_0)T_0/(2\hbar) ε = q 2 E ′′ ( p 0 ) T 0 / ( 2ℏ ) . In the slow-electron BR dispersion E ′ ′ = 1 / m E''=1/m E ′′ = 1/ m . No unspecified reference frequency or new coupling is needed.
After removal of the common initial free phase, the Raman–Nath ladder is
i d C ℓ d τ = ( − W ℓ + ε ℓ 2 ) C ℓ + g [ n + ℓ + 1 C ℓ + 1 + n + ℓ C ℓ − 1 ] . (47) i\frac{dC_{\ell}}{d\tau}=(-W\ell+\varepsilon\ell^2)C_{\ell}+g\left[\sqrt{n+\ell+1}\,C_{\ell+1}+\sqrt{n+\ell}\,C_{\ell-1}\right]. \tag*{(47)} i d τ d C ℓ = ( − W ℓ + ε ℓ 2 ) C ℓ + g [ n + ℓ + 1 C ℓ + 1 + n + ℓ C ℓ − 1 ] . (47)
The boundary n + ℓ ≥ 0 n+\ell\geq0 n + ℓ ≥ 0 follows from photon-number positivity. The square-root factors encode stimulated and vacuum emission; the diagonal terms encode detuning and recoil. The Taylor reduction of the electron energy remains controlled only while the occupied momentum shifts stay in its validity range. Higher recoil orders are not automatically described accurately by the quadratic approximation.
For the conditional vacuum input n = 0 n=0 n = 0 , initially sharp electron momentum and initial condition Eq. (44) used in the following perturbative calculation, remove the diagonal phase exactly:
C ℓ ( τ ) = e − i ( − W ℓ + ε ℓ 2 ) τ D ℓ ( τ ) . (48) C_\ell(\tau)=e^{-i(-W\ell+\varepsilon\ell^2)\tau}D_\ell(\tau). \tag*{(48)} C ℓ ( τ ) = e − i ( − W ℓ + ε ℓ 2 ) τ D ℓ ( τ ) . (48)
Substitution into Eq. (47) gives
i D ˙ ℓ = g ℓ + 1 e i W τ − i ε ( 2 ℓ + 1 ) τ D ℓ + 1 + g ℓ e − i W τ + i ε ( 2 ℓ − 1 ) τ D ℓ − 1 . (49) i\dot{D}_\ell=g\sqrt{\ell+1}e^{iW\tau-i\varepsilon(2\ell+1)\tau}D_{\ell+1}+g\sqrt{\ell}e^{-iW\tau+i\varepsilon(2\ell-1)\tau}D_{\ell-1}. \tag*{(49)} i D ˙ ℓ = g ℓ + 1 e iW τ − i ε ( 2 ℓ + 1 ) τ D ℓ + 1 + g ℓ e − iW τ + i ε ( 2 ℓ − 1 ) τ D ℓ − 1 . (49)
The dot denotes differentiation with respect to τ \tau τ . The neighboring diagonal differences are
d ℓ + 1 − d ℓ = − W + ε ( 2 ℓ + 1 ) , d ℓ − d ℓ − 1 = − W + ε ( 2 ℓ − 1 ) , d ℓ = − W ℓ + ε ℓ 2 . (50) d_{\ell+1}-d_\ell=-W+\varepsilon(2\ell+1),\qquad d_\ell-d_{\ell-1}=-W+\varepsilon(2\ell-1),\qquad d_\ell=-W\ell+\varepsilon\ell^2. \tag*{(50)} d ℓ + 1 − d ℓ = − W + ε ( 2 ℓ + 1 ) , d ℓ − d ℓ − 1 = − W + ε ( 2 ℓ − 1 ) , d ℓ = − W ℓ + ε ℓ 2 . (50)
The local mismatch means the energy difference between two neighboring exchange states, divided by the reference frequency 1 / T 0 1/T_0 1/ T 0 , including their detuning. This difference represents the local mismatch and fixes the phase of that transition. The local mismatch, rather than the absolute diagonal energy ε ℓ 2 \varepsilon\ell^2 ε ℓ 2 , controls the transition phases. This is the physical reason for expanding the transformed couplings. We assume that
∣ ε ( 2 ℓ ± 1 ) τ ∣ ≪ 1 (51) |\varepsilon(2\ell\pm1)\tau|\ll1 \tag*{(51)} ∣ ε ( 2 ℓ ± 1 ) τ ∣ ≪ 1 (51)
throughout the significantly populated ladder. Keeping W W W nonzero and exact, the phase factors become
e i W τ − i ε ( 2 ℓ + 1 ) τ = e i W τ [ 1 − i ε ( 2 ℓ + 1 ) τ − ε 2 2 ( 2 ℓ + 1 ) 2 τ 2 ] + O ( ε 3 ) , e − i W τ + i ε ( 2 ℓ − 1 ) τ = e − i W τ [ 1 + i ε ( 2 ℓ − 1 ) τ − ε 2 2 ( 2 ℓ − 1 ) 2 τ 2 ] + O ( ε 3 ) . (52) \begin{aligned}
e^{iW\tau-i\varepsilon(2\ell+1)\tau}&=e^{iW\tau}\left[1-i\varepsilon(2\ell+1)\tau-\frac{\varepsilon^2}{2}(2\ell+1)^2\tau^2\right]+O(\varepsilon^3),\\
e^{-iW\tau+i\varepsilon(2\ell-1)\tau}&=e^{-iW\tau}\left[1+i\varepsilon(2\ell-1)\tau-\frac{\varepsilon^2}{2}(2\ell-1)^2\tau^2\right]+O(\varepsilon^3). \tag*{(52)}
\end{aligned} e iW τ − i ε ( 2 ℓ + 1 ) τ e − iW τ + i ε ( 2 ℓ − 1 ) τ = e iW τ [ 1 − i ε ( 2 ℓ + 1 ) τ − 2 ε 2 ( 2 ℓ + 1 ) 2 τ 2 ] + O ( ε 3 ) , = e − iW τ [ 1 + i ε ( 2 ℓ − 1 ) τ − 2 ε 2 ( 2 ℓ − 1 ) 2 τ 2 ] + O ( ε 3 ) . (52)
Writing
D ℓ = D ℓ ( 0 ) + ε D ℓ ( 1 ) + ε 2 D ℓ ( 2 ) + O ( ε 3 ) , (53) D_\ell=D_\ell^{(0)}+\varepsilon D_\ell^{(1)}+\varepsilon^2D_\ell^{(2)}+O(\varepsilon^3), \tag*{(53)} D ℓ = D ℓ ( 0 ) + ε D ℓ ( 1 ) + ε 2 D ℓ ( 2 ) + O ( ε 3 ) , (53)
introduce the exact recoilless, detuned ladder operator
( H 0 D ) ℓ = g ( e i W τ ℓ + 1 D ℓ + 1 + e − i W τ ℓ D ℓ − 1 ) . (54) (H_0D)_\ell=g\left(e^{iW\tau}\sqrt{\ell+1}D_{\ell+1}+e^{-iW\tau}\sqrt{\ell}D_{\ell-1}\right). \tag*{(54)} ( H 0 D ) ℓ = g ( e iW τ ℓ + 1 D ℓ + 1 + e − iW τ ℓ D ℓ − 1 ) . (54)
The various orders of the perturbative expansion are linked by the differential equations
i D ˙ ( 0 ) = H 0 D ( 0 ) , i D ˙ ( 1 ) = H 0 D ( 1 ) + S ( 1 ) , i D ˙ ( 2 ) = H 0 D ( 2 ) + S ( 2 ) , (55) i\dot{D}^{(0)}=H_0D^{(0)},\qquad i\dot{D}^{(1)}=H_0D^{(1)}+S^{(1)},\qquad i\dot{D}^{(2)}=H_0D^{(2)}+S^{(2)}, \tag*{(55)} i D ˙ ( 0 ) = H 0 D ( 0 ) , i D ˙ ( 1 ) = H 0 D ( 1 ) + S ( 1 ) , i D ˙ ( 2 ) = H 0 D ( 2 ) + S ( 2 ) , (55)
where
S ℓ ( 1 ) = − i g τ ( 2 ℓ + 1 ) e i W τ ℓ + 1 D ℓ + 1 ( 0 ) + i g τ ( 2 ℓ − 1 ) e − i W τ ℓ D ℓ − 1 ( 0 ) , (56) S_\ell^{(1)}=-ig\tau(2\ell+1)e^{iW\tau}\sqrt{\ell+1}D_{\ell+1}^{(0)}+ig\tau(2\ell-1)e^{-iW\tau}\sqrt{\ell}D_{\ell-1}^{(0)}, \tag*{(56)} S ℓ ( 1 ) = − i g τ ( 2 ℓ + 1 ) e iW τ ℓ + 1 D ℓ + 1 ( 0 ) + i g τ ( 2 ℓ − 1 ) e − iW τ ℓ D ℓ − 1 ( 0 ) , (56)
and
S ℓ ( 2 ) = − i g τ ( 2 ℓ + 1 ) e i W τ ℓ + 1 D ℓ + 1 ( 1 ) + i g τ ( 2 ℓ − 1 ) e − i W τ ℓ D ℓ − 1 ( 1 ) − g τ 2 2 ( 2 ℓ + 1 ) 2 e i W τ ℓ + 1 D ℓ + 1 ( 0 ) − g τ 2 2 ( 2 ℓ − 1 ) 2 e − i W τ ℓ D ℓ − 1 ( 0 ) . (57) S_\ell^{(2)}=-ig\tau(2\ell+1)e^{iW\tau}\sqrt{\ell+1}D_{\ell+1}^{(1)}+ig\tau(2\ell-1)e^{-iW\tau}\sqrt{\ell}D_{\ell-1}^{(1)}-\frac{g\tau^2}{2}(2\ell+1)^2e^{iW\tau}\sqrt{\ell+1}D_{\ell+1}^{(0)}-\frac{g\tau^2}{2}(2\ell-1)^2e^{-iW\tau}\sqrt{\ell}D_{\ell-1}^{(0)}. \tag*{(57)} S ℓ ( 2 ) = − i g τ ( 2 ℓ + 1 ) e iW τ ℓ + 1 D ℓ + 1 ( 1 ) + i g τ ( 2 ℓ − 1 ) e − iW τ ℓ D ℓ − 1 ( 1 ) − 2 g τ 2 ( 2 ℓ + 1 ) 2 e iW τ ℓ + 1 D ℓ + 1 ( 0 ) − 2 g τ 2 ( 2 ℓ − 1 ) 2 e − iW τ ℓ D ℓ − 1 ( 0 ) . (57)
If U 0 ( τ , s ) U_0(\tau,s) U 0 ( τ , s ) is the propagator of Eq. (54) , and the initial state has no recoil-dependent correction, then
D ( j ) ( τ ) = − i ∫ 0 τ U 0 ( τ , s ) S ( j ) ( s ) d s , j = 1 , 2. (58) D^{(j)}(\tau)=-i\int_0^\tau U_0(\tau,s)S^{(j)}(s)\,\mathrm{d}s,\qquad j=1,2. \tag*{(58)} D ( j ) ( τ ) = − i ∫ 0 τ U 0 ( τ , s ) S ( j ) ( s ) d s , j = 1 , 2. (58)
Equation (58) is the Duhamel formula, or variation-of-constants solution [34 ] , for i ∂ τ D ( j ) = H 0 ( τ ) D ( j ) + S ( j ) ( τ ) i\partial_\tau D^{(j)}=H_0(\tau)D^{(j)}+S^{(j)}(\tau) i ∂ τ D ( j ) = H 0 ( τ ) D ( j ) + S ( j ) ( τ ) with D ( j ) ( 0 ) = 0 D^{(j)}(0)=0 D ( j ) ( 0 ) = 0 . Here U 0 ( τ , s ) U_0(\tau,s) U 0 ( τ , s ) satisfies i ∂ τ U 0 = H 0 ( τ ) U 0 i\partial_\tau U_0=H_0(\tau)U_0 i ∂ τ U 0 = H 0 ( τ ) U 0 and U 0 ( s , s ) = I U_0(s,s)=I U 0 ( s , s ) = I . Multiplying by U 0 ( s , τ ) U_0(s,\tau) U 0 ( s , τ ) and integrating yields the formula; differentiating it recovers the inhomogeneous equation and its initial condition. It is an exact identity at each hierarchy order. Its iteration is the time-ordered Dyson construction [35 , 36 ] ; only the subsequent finite-order truncation is approximate. The “Duhamel integrals” below refer specifically to these source-propagation integrals.
These equations give an operational expansion: detuning is retained exactly, while recoil enters through local mismatch and explicit source terms.
For one elementary neighboring transition, define the dimensionless frequency argument ν \nu ν and its finite-time integral
A ( ν , τ ) = ∫ 0 τ e i ν s d s = τ e i ν τ / 2 sinc ( ν τ / 2 ) , S 0 ( W , τ ) = τ 2 sinc 2 ( W τ / 2 ) . (59) \mathcal{A}(\nu,\tau)=\int_0^\tau e^{i\nu s}\,\mathrm{d}s=\tau e^{i\nu\tau/2}\operatorname{sinc}(\nu\tau/2),\qquad S_0(W,\tau)=\tau^2\operatorname{sinc}^2(W\tau/2). \tag*{(59)} A ( ν , τ ) = ∫ 0 τ e i ν s d s = τ e i ν τ /2 sinc ( ν τ /2 ) , S 0 ( W , τ ) = τ 2 sinc 2 ( W τ /2 ) . (59)
We use sinc x = sin x / x \operatorname{sinc}x=\sin x/x sinc x = sin x / x with sinc 0 = 1 \operatorname{sinc}0=1 sinc 0 = 1 . These forms are explicitly regular: A ( 0 , τ ) = τ \mathcal{A}(0,\tau)=\tau A ( 0 , τ ) = τ and S 0 ( 0 , τ ) = τ 2 S_0(0,\tau)=\tau^2 S 0 ( 0 , τ ) = τ 2 . Any denominator in the equivalent trigonometric expressions is therefore a removable singularity, not a physical divergence. A local recoil shift δ ℓ ( ± ) = ε ( 2 ℓ ± 1 ) \delta_\ell^{(\pm)}=\varepsilon(2\ell\pm1) δ ℓ ( ± ) = ε ( 2 ℓ ± 1 ) yields
S 0 ( W − δ , τ ) = S 0 + δ G 1 + δ 2 G 2 + O ( δ 3 ) , (60) S_0(W-\delta,\tau)=S_0+\delta G_1+\delta^2G_2+O(\delta^3), \tag*{(60)} S 0 ( W − δ , τ ) = S 0 + δ G 1 + δ 2 G 2 + O ( δ 3 ) , (60)
with the explicit functions
G 1 ( W , τ ) = − ∂ W S 0 = 4 W 3 [ 1 − cos ( W τ ) − W τ 2 sin ( W τ ) ] , (61) G_1(W,\tau)=-\partial_W S_0=\frac{4}{W^3}\left[1-\cos(W\tau)-\frac{W\tau}{2}\sin(W\tau)\right], \tag*{(61)} G 1 ( W , τ ) = − ∂ W S 0 = W 3 4 [ 1 − cos ( W τ ) − 2 W τ sin ( W τ ) ] , (61)
G 2 ( W , τ ) = 1 2 ∂ W 2 S 0 = W 2 τ 2 cos ( W τ ) − 4 W τ sin ( W τ ) − 6 cos ( W τ ) + 6 W 4 . (62) G_2(W,\tau)=\frac{1}{2}\partial_W^2S_0=\frac{W^2\tau^2\cos(W\tau)-4W\tau\sin(W\tau)-6\cos(W\tau)+6}{W^4}. \tag*{(62)} G 2 ( W , τ ) = 2 1 ∂ W 2 S 0 = W 4 W 2 τ 2 cos ( W τ ) − 4 W τ sin ( W τ ) − 6 cos ( W τ ) + 6 . (62)
The apparent inverse powers of W W W in these derivatives also cancel. Expanding the entire line shape about W = 0 W=0 W = 0 gives
S 0 = τ 2 − W 2 τ 4 12 + W 4 τ 6 360 + O ( W 6 τ 8 ) , G 1 = W τ 4 6 − W 3 τ 6 90 + O ( W 5 τ 8 ) , G 2 = − τ 4 12 + W 2 τ 6 60 + O ( W 4 τ 8 ) . (63) \begin{aligned}
S_0&=\tau^2-\frac{W^2\tau^4}{12}+\frac{W^4\tau^6}{360}+O(W^6\tau^8),\\
G_1&=\frac{W\tau^4}{6}-\frac{W^3\tau^6}{90}+O(W^5\tau^8),\\
G_2&=-\frac{\tau^4}{12}+\frac{W^2\tau^6}{60}+O(W^4\tau^8). \tag*{(63)}
\end{aligned} S 0 G 1 G 2 = τ 2 − 12 W 2 τ 4 + 360 W 4 τ 6 + O ( W 6 τ 8 ) , = 6 W τ 4 − 90 W 3 τ 6 + O ( W 5 τ 8 ) , = − 12 τ 4 + 60 W 2 τ 6 + O ( W 4 τ 8 ) . (63)
Thus G 1 ( 0 , τ ) = 0 G_1(0,\tau)=0 G 1 ( 0 , τ ) = 0 and G 2 ( 0 , τ ) = − τ 4 / 12 G_2(0,\tau)=-\tau^4/12 G 2 ( 0 , τ ) = − τ 4 /12 . These series, or the sinc representation and its derivatives, should be used numerically near zero detuning to avoid cancellation errors. The amplitude hierarchy is itself regular there: its source terms contain only polynomials in τ \tau τ and exponentials e ± i W τ e^{\pm iW\tau} e ± iW τ . On a finite ladder the Duhamel integrals have no zero-detuning singularity. Their use on the infinite ladder additionally requires the population and phase control discussed in Section VI.
At τ = 1 \tau=1 τ = 1 , G 1 ( W , 1 ) = − ∂ W sinc 2 ( W / 2 ) G_1(W,1)=-\partial_W\operatorname{sinc}^2(W/2) G 1 ( W , 1 ) = − ∂ W sinc 2 ( W /2 ) is the usual antisymmetric low-gain FEL function, apart from the conventional multiplicative gain coefficient. In the normalization of Ref. [37 ] , the space-charge-free small-signal gain is G ( W , 0 ) = π g 0 G 1 ( W , 1 ) G(W,0)=\pi g_0G_1(W,1) G ( W , 0 ) = π g 0 G 1 ( W , 1 ) . The frequency orientation must be matched when comparing signs. This identifies the physical meaning of the first derivative without introducing a separate complex-response hierarchy.
More generally G j = ( − 1 ) j ∂ W j S 0 / j ! G_j=(-1)^j\partial_W^jS_0/j! G j = ( − 1 ) j ∂ W j S 0 / j ! . The first derivative is antisymmetric in W W W , and the curvature correction is even. Locally each rung samples a shifted line shape; the full probability still requires the coupled amplitude evolution in Eq. (55) , rather than a separate replacement of the mean in every probability.
For an initially sharp electron momentum and n n n initial photons, first-order transition amplitudes give, to second order in the coupling,
P e m ≃ g 2 ( n + 1 ) S 0 ( W − ε , τ ) , P a b s ≃ g 2 n S 0 ( W + ε , τ ) . (64) P_{\rm em} \simeq g^{2}(n+1)S_{0}(W-\varepsilon,\tau), \qquad P_{\rm abs} \simeq g^{2}nS_{0}(W+\varepsilon,\tau). \tag*{(64)} P em ≃ g 2 ( n + 1 ) S 0 ( W − ε , τ ) , P abs ≃ g 2 n S 0 ( W + ε , τ ) . (64)
The stimulated photon change is consequently
Δ n s t i m ≃ n g 2 [ S 0 ( W − ε , τ ) − S 0 ( W + ε , τ ) ] = 2 n g 2 ε G 1 ( W , τ ) + O ( ε 3 ) . (65) \Delta n_{\rm stim} \simeq ng^{2}[S_{0}(W-\varepsilon,\tau)-S_{0}(W+\varepsilon,\tau)] = 2ng^{2}\varepsilon\mathcal{G}_{1}(W,\tau)+O(\varepsilon^{3}). \tag*{(65)} Δ n stim ≃ n g 2 [ S 0 ( W − ε , τ ) − S 0 ( W + ε , τ )] = 2 n g 2 ε G 1 ( W , τ ) + O ( ε 3 ) . (65)
Equation (65) is evaluated to leading order g 2 g^{2} g 2 in the transition probabilities. The displayed remainder refers to the recoil expansion at that order; higher orders in the coupling are also neglected.
The remaining g 2 S 0 ( W − ε , τ ) g^{2}S_{0}(W-\varepsilon,\tau) g 2 S 0 ( W − ε , τ ) term is spontaneous feeding of the selected mode. This explicitly connects the n + 1 n+1 n + 1 factor, recoil shifts and the derivative structure of the small-signal gain [1 , 7 ] . The displayed result is a weak-transition single-electron expression, not a self-consistent high-gain growth law.
We checked the amplitude hierarchy directly against propagation of the time-independent Raman–Nath Hamiltonian. For the vacuum initial condition, g = 0.7 g=0.7 g = 0.7 and τ = 1 \tau=1 τ = 1 , let
E 2 = ∥ C e x a c t − e − i d ℓ τ ( D ( 0 ) + ε D ( 1 ) + ε 2 D ( 2 ) ) ∥ 2 , d ℓ = − W ℓ + ε ℓ 2 . (66) \mathcal{E}_{2}=\left\|C_{\rm exact}-e^{-id_{\ell}\tau}\left(D^{(0)}+\varepsilon D^{(1)}+\varepsilon^{2}D^{(2)}\right)\right\|_{2}, \qquad d_{\ell}=-W\ell+\varepsilon\ell^{2}. \tag*{(66)} E 2 = C exact − e − i d ℓ τ ( D ( 0 ) + ε D ( 1 ) + ε 2 D ( 2 ) ) 2 , d ℓ = − W ℓ + ε ℓ 2 . (66)
The phase multiplication in this expression is componentwise. Matrix-exponential propagation supplies C e x a c t C_{\rm exact} C exact ; numerical integration of Eq. (55) supplies the three perturbative coefficients. The results are
Table 1.
Increasing the truncation from 0 ≤ ℓ ≤ 35 0 \leq\ell\leq35 0 ≤ ℓ ≤ 35 to 0 ≤ ℓ ≤ 47 0 \leq\ell\leq47 0 ≤ ℓ ≤ 47 changes the reported errors by less than 10 − 12 10^{-12} 1 0 − 12 . The hierarchy was integrated with relative tolerance 2 × 10 − 12 2 \times10^{-12} 2 × 1 0 − 12 and absolute tolerance 2 × 10 − 14 2 \times10^{-14} 2 × 1 0 − 14 . Halving ε \varepsilon ε reduces the error by approximately eight, consistent with an omitted cubic amplitude correction, including at W = 0 W=0 W = 0 . At ε = 0 \varepsilon=0 ε = 0 the numerical photon probabilities agree with the analytic Poisson law to within 4 × 10 − 15 4 \times10^{-15} 4 × 1 0 − 15 . Here the normalization residuals mean ∣ ∑ ℓ ∣ D ℓ ( 0 ) ∣ 2 − 1 ∣ \left|\sum_{\ell}|D_{\ell}^{(0)}|^{2}-1\right| ∑ ℓ ∣ D ℓ ( 0 ) ∣ 2 − 1 ,
∣ ∑ ℓ 2 Re ( D ℓ ( 0 ) ∗ D ℓ ( 1 ) ) ∣ \left|\sum_{\ell}2\operatorname{Re}(D_{\ell}^{(0)*}D_{\ell}^{(1)})\right| ∑ ℓ 2 Re ( D ℓ ( 0 ) ∗ D ℓ ( 1 ) ) and ∣ ∑ ℓ [ ∣ D ℓ ( 1 ) ∣ 2 + 2 Re ( D ℓ ( 0 ) ∗ D ℓ ( 2 ) ) ] ∣ \left|\sum_{\ell}[|D_{\ell}^{(1)}|^{2}+2\operatorname{Re}(D_{\ell}^{(0)*}D_{\ell}^{(2)})]\right| ∑ ℓ [ ∣ D ℓ ( 1 ) ∣ 2 + 2 Re ( D ℓ ( 0 ) ∗ D ℓ ( 2 ) )] . All three are below 5 × 10 − 15 5 \times10^{-15} 5 × 1 0 − 15 in these tests. These checks show that the terms of order 1 1 1 , ε \varepsilon ε and ε 2 \varepsilon^{2} ε 2 in the norm have the required values. At nonzero ε \varepsilon ε , the truncated state can still have a norm error from terms of order ε 3 \varepsilon^{3} ε 3 and higher; that error must be checked separately. These are checks of the conditional ladder calculation in this parameter range, not a claim of uniform accuracy at arbitrary coupling, time or occupied recoil order.
Photon statistics and FEL-induced energy spread
The earlier many-particle analysis of Becker and McIver [9 ] uses an interaction-picture recoil expansion and treats photon fluctuations as well as mean gain. Their review [10 ] places photon statistics and cooperative effects at the centre of the quantum description. This historical connection is important: non-Poissonian startup statistics are not by themselves a unique signature of a recoil-resolved oscillator, and an observable prediction must retain the relevant preparation and beam averaging.
In the preceding calculation we treated the vacuum-input RN equation, n = 0 n=0 n = 0 . Its ℓ + 1 \sqrt{\ell+1} ℓ + 1 and ℓ \sqrt{\ell} ℓ factors originate from photon creation and annihilation. At zero recoil its solution has Poisson probability amplitudes. Within a fixed initial photon-number sector, for a strongly populated mode with n ≫ ∣ ℓ ∣ n \gg|\ell| n ≫ ∣ ℓ ∣ over every populated exchange order, both square roots can instead be approximated by n \sqrt{n} n and absorbed into the coupling h = g n h=g\sqrt{n} h = g n . The equation then becomes the shift RN equation [36 ] , related to the original ultrasound diffraction problem [38 ] :
i C ˙ ℓ = ( − W ℓ + ε ℓ 2 ) C ℓ + h ( C ℓ + 1 + C ℓ − 1 ) . (67) i\dot{C}_{\ell} = (-W\ell+ \varepsilon\ell^{2})C_{\ell} + h(C_{\ell+1} + C_{\ell-1}). \tag*{(67)} i C ˙ ℓ = ( − W ℓ + ε ℓ 2 ) C ℓ + h ( C ℓ + 1 + C ℓ − 1 ) . (67)
The lower photon boundary is remote in this approximation. At ε = 0 \varepsilon= 0 ε = 0 , for C ℓ ( 0 ) = δ ℓ 0 C_{\ell}(0) = \delta_{\ell0} C ℓ ( 0 ) = δ ℓ 0 and effectively integer ℓ \ell ℓ of either sign,
C ℓ ( 0 ) ( τ ) = ( − i ) ℓ e i ℓ W τ / 2 J ℓ [ 2 h τ sinc ( W τ / 2 ) ] . (68) C_{\ell}^{(0)}(\tau) = (-i)^{\ell}e^{i\ell W\tau/2}J_{\ell}[2h\tau\operatorname{sinc}(W\tau/2)]. \tag*{(68)} C ℓ ( 0 ) ( τ ) = ( − i ) ℓ e i ℓ W τ /2 J ℓ [ 2 h τ sinc ( W τ /2 )] . (68)
Because J − ℓ 2 = J ℓ 2 J_{-\ell}^{2} = J_{\ell}^{2} J − ℓ 2 = J ℓ 2 , the fixed-number exchange probabilities are symmetric and ⟨ ℓ ⟩ = 0 \langle\ell\rangle= 0 ⟨ ℓ ⟩ = 0 in this recoil-free approximation. This is an exchange-order distribution referenced to ∣ n ⟩ \lvert n\rangle ∣ n ⟩ , not the complete photon-count distribution of a coherently seeded laser. The Poisson and Bessel formulas do not identify two distinct physical recoil regimes. They are representations or approximations within the same recoil-free dynamics, evaluated with different input-state specifications. For a coherent seed, the number-sector amplitudes in Eq. (45) must be superposed with c n = e − ∣ α 0 ∣ 2 / 2 α 0 n / n ! c_n=e^{-|\alpha_0|^2/2}\alpha_0^n/\sqrt{n!} c n = e − ∣ α 0 ∣ 2 /2 α 0 n / n ! before constructing the final density operator. Replacing the coherent seed by ∣ n ⟩ \lvert n\rangle ∣ n ⟩ , or by an incoherent average of number-state exchange probabilities, removes information required for phase-sensitive observables.
The common coherent-state structure is explicit in the prescribed-current reference. Let β \beta β be the displacement generated by the source in the radiation interaction picture. Then
U I = e i Φ D ( β ) , D ( β ) ∣ α 0 ⟩ = e ( β α 0 ∗ − β ∗ α 0 ) / 2 ∣ α 0 + β ⟩ , D ( β ) = e β a ^ † − β ∗ a ^ . (69) U_I=e^{i\Phi}D(\beta),\quad D(\beta)\lvert\alpha_0\rangle=e^{(\beta\alpha_0^*-\beta^*\alpha_0)/2}\lvert\alpha_0+\beta\rangle,\quad D(\beta)=e^{\beta\hat{a}^{\dagger}-\beta^*\hat{a}}. \tag*{(69)} U I = e i Φ D ( β ) , D ( β ) ∣ α 0 ⟩ = e ( β α 0 ∗ − β ∗ α 0 ) /2 ∣ α 0 + β ⟩ , D ( β ) = e β a ^ † − β ∗ a ^ . (69)
Vacuum is the case α 0 = 0 \alpha_0=0 α 0 = 0 ; a nonzero coherent seed obeys the same displacement evolution. At a fixed source and optical phase both give a Poisson law with mean ∣ α 0 + β ∣ 2 \lvert\alpha_0+\beta\rvert^2 ∣ α 0 + β ∣ 2 . A Bessel exchange kernel in a fixed-number approximation cannot be substituted for that law. Phase-dependent interference can change the field energy for a coherent seed, so the vanishing mean exchange in the fixed-number calculation must not be transferred indiscriminately to it.
When the electron is retained as a quantum degree of freedom, the zero-recoil displacement involves the dressed operator A ^ \hat{A} A ^ , not simply the bare radiation operator. The reduced optical statistics can then depend on electron coherence and tracing; a coherent optical seed alone does not ensure an unentangled coherent output. The appropriate comparison therefore keeps the same Hamiltonian and the stated electron and optical preparation. The point made in Ref. [39 ] is more specific: treating the undulator pump as a photon-number state produces antibunching even at zero electron recoil. Such a state does not represent the coherent classical drive intended in that calculation. The coherent pump removes this artifact. This concerns the undulator mode, not merely the choice of a radiation seed.
To display the result, retain both modes in Eq. (9) , take zero initial radiation photons and zero recoil, and let m m m be the initial pump occupation. The exact spherical Raman–Nath solution in Ref. [39 ] gives
P ( ℓ ∣ m ) = ( m ℓ ) χ ℓ ( 1 − χ ) m − ℓ , χ = 4 g u 2 4 g u 2 + W 2 sin 2 [ τ 2 4 g u 2 + W 2 ] , g u = g u 0 . (70) P(\ell\mid m)=\binom{m}{\ell}\chi^{\ell}(1-\chi)^{m-\ell},\quad\chi=\frac{4g_u^2}{4g_u^2+W^2}\sin^2\left[\frac{\tau}{2}\sqrt{4g_u^2+W^2}\right],\quad g_u=g_{u0}. \tag*{(70)} P ( ℓ ∣ m ) = ( ℓ m ) χ ℓ ( 1 − χ ) m − ℓ , χ = 4 g u 2 + W 2 4 g u 2 sin 2 [ 2 τ 4 g u 2 + W 2 ] , g u = g u 0 . (70)
Here 0 ≤ ℓ ≤ m 0\leq\ell\leq m 0 ≤ ℓ ≤ m counts transferred pump quanta and emitted radiation photons, and χ \chi χ is the single-quantum transfer probability. The phase of the amplitudes is needed for coherence observables but not for this population law. A pump Fock state has
⟨ ℓ ⟩ m = m χ , Var ( ℓ ∣ m ) = m χ ( 1 − χ ) , Q M = Var ( ℓ ) − ⟨ ℓ ⟩ ⟨ ℓ ⟩ = − χ , (71) \langle\ell\rangle_m=m\chi,\quad\operatorname{Var}(\ell\mid m)=m\chi(1-\chi),\quad Q_M=\frac{\operatorname{Var}(\ell)-\langle\ell\rangle}{\langle\ell\rangle}=-\chi, \tag*{(71)} ⟨ ℓ ⟩ m = m χ , Var ( ℓ ∣ m ) = m χ ( 1 − χ ) , Q M = ⟨ ℓ ⟩ Var ( ℓ ) − ⟨ ℓ ⟩ = − χ , (71)
for nonzero mean. The symbol Q M Q_M Q M denotes the Mandel parameter, distinct from cavity quality factor or bunch charge. The negative value is present without recoil; it is inherited from the number-state pump.
For a coherent undulator input ∣ β u ⟩ \lvert\beta_u\rangle ∣ β u ⟩ , let m ˉ = ∣ β u ∣ 2 \bar{m}=\lvert\beta_u\rvert^2 m ˉ = ∣ β u ∣ 2 . Averaging the number populations with the coherent-state weights yields
P ℓ = ∑ m = ℓ ∞ e − m ˉ m ˉ m m ! ( m ℓ ) χ ℓ ( 1 − χ ) m − ℓ = e − χ m ˉ ( χ m ˉ ) ℓ ℓ ! , P_{\ell}=\sum_{m=\ell}^{\infty}e^{-\bar{m}}\frac{\bar{m}^{m}}{m!}\binom{m}{\ell}\chi^{\ell}(1-\chi)^{m-\ell}=e^{-\chi\bar{m}}\frac{(\chi\bar{m})^{\ell}}{\ell!}, P ℓ = m = ℓ ∑ ∞ e − m ˉ m ! m ˉ m ( ℓ m ) χ ℓ ( 1 − χ ) m − ℓ = e − χ m ˉ ℓ ! ( χ m ˉ ) ℓ ,
⟨ ℓ ⟩ = χ m ˉ , Var ( ℓ ) = χ m ˉ , Q M = 0. (72) \langle\ell\rangle=\chi\bar{m},\quad\operatorname{Var}(\ell)=\chi\bar{m},\quad Q_M=0. \tag*{(72)} ⟨ ℓ ⟩ = χ m ˉ , Var ( ℓ ) = χ m ˉ , Q M = 0. (72)
This is the explicit correction to the artificial zero-recoil antibunching. The counting calculation depends on the pump number weights; a phase-random state with those same Poisson weights gives the same populations but not the same phase coherence. The coherent-pump amplitudes are required for the coherent Bloch-state interpretation used in the reference. Poisson counts alone still do not prove a pure coherent reduced radiation state after tracing out the electron. In the large-pump, weak-depletion limit, χ m ˉ → g 2 S 0 \chi_{\bar{m}} \to g^2 S_0 χ m ˉ → g 2 S 0 with g = g u m ˉ g = g_u \sqrt{\bar{m}} g = g u m ˉ , recovering the one-mode Poisson reference below. Thus the appropriate recoil test first specifies the pump preparation and then compares finite-recoil predictions with this corrected zero-recoil baseline.
Before proceeding, we underline the role of A ^ = a ^ T ^ q † \hat{A} = \hat{a}\hat{T}_q^\dagger A ^ = a ^ T ^ q † , the dressed annihilation operator. At zero recoil the vacuum exchange-ladder state evolves into a coherent state in the Glauber sense of this joint electron–field operator. Kinetic recoil breaks that coherence-preserving dynamical structure and generally produces a state that is not an eigenstate of A ^ \hat{A} A ^ , together with non-Poissonian probabilities. This is joint coherence; it does not by itself make the reduced optical state a pure coherent state. The operator algebra [ A ^ , A ^ † ] = 1 [\hat{A},\hat{A}^\dagger] = 1 [ A ^ , A ^ † ] = 1 remains valid even when recoil is present; it is the evolution’s coherent-state preservation that fails.
The diagonal transformation does not change probabilities:
P ℓ = ∣ C ℓ ∣ 2 = ∣ D ℓ ∣ 2 = P ℓ ( 0 ) + ε P ℓ ( 1 ) + ε 2 P ℓ ( 2 ) + O ( ε 3 ) , (73) P_\ell= |C_\ell|^2 = |D_\ell|^2 = P_\ell^{(0)} + \varepsilon P_\ell^{(1)} + \varepsilon^2 P_\ell^{(2)} + O(\varepsilon^3), \tag*{(73)} P ℓ = ∣ C ℓ ∣ 2 = ∣ D ℓ ∣ 2 = P ℓ ( 0 ) + ε P ℓ ( 1 ) + ε 2 P ℓ ( 2 ) + O ( ε 3 ) , (73)
where
P ℓ ( 1 ) = 2 Re [ D ℓ ( 0 ) ∗ D ℓ ( 1 ) ] , P ℓ ( 2 ) = ∣ D ℓ ( 1 ) ∣ 2 + 2 Re [ D ℓ ( 0 ) ∗ D ℓ ( 2 ) ] . (74) P_\ell^{(1)} = 2\operatorname{Re}[D_\ell^{(0)*}D_\ell^{(1)}], \qquad P_\ell^{(2)} = |D_\ell^{(1)}|^2 + 2\operatorname{Re}[D_\ell^{(0)*}D_\ell^{(2)}]. \tag*{(74)} P ℓ ( 1 ) = 2 Re [ D ℓ ( 0 ) ∗ D ℓ ( 1 ) ] , P ℓ ( 2 ) = ∣ D ℓ ( 1 ) ∣ 2 + 2 Re [ D ℓ ( 0 ) ∗ D ℓ ( 2 ) ] . (74)
At order ε 2 \varepsilon^2 ε 2 , two contributions arise when the amplitude is squared: the product of the first-order correction with its conjugate, and the interference of the zeroth- and second-order amplitudes. Both must be retained for a consistent second-order probability.
For the conditional preparation ∣ p 0 ⟩ ⊗ ∣ 0 ⟩ |p_0\rangle\otimes|0\rangle ∣ p 0 ⟩ ⊗ ∣0 ⟩ with C ℓ ( 0 ) = δ ℓ 0 C_\ell(0) = \delta_{\ell0} C ℓ ( 0 ) = δ ℓ 0 , at ε = 0 \varepsilon= 0 ε = 0 the vacuum ladder amplitudes have the displaced-oscillator form, yielding
μ ( W , τ ) = g 2 S 0 ( W , τ ) , P ℓ ( 0 ) = e − μ μ ℓ ℓ ! , F ( 0 ) = 1 , g 0 ( 2 ) ( 0 ) = 1. (75) \mu(W,\tau) = g^2 S_0(W,\tau), \qquad P_\ell^{(0)} = e^{-\mu}\frac{\mu^\ell}{\ell!}, \qquad F^{(0)} = 1, \qquad g_0^{(2)}(0) = 1. \tag*{(75)} μ ( W , τ ) = g 2 S 0 ( W , τ ) , P ℓ ( 0 ) = e − μ ℓ ! μ ℓ , F ( 0 ) = 1 , g 0 ( 2 ) ( 0 ) = 1. (75)
Here “conditional” means calculated at a fixed initial electron momentum p p p and specified initial photon number, before averaging over the entrance beam distribution; it does not imply detector postselection. The variance is explicitly
Var ( n ) = ⟨ ( n − ⟨ n ⟩ ) 2 ⟩ = ∑ n n 2 P n − ( ∑ n n P n ) 2 . (76) \operatorname{Var}(n) = \langle(n-\langle n\rangle)^2\rangle= \sum_n n^2P_n - \left(\sum_n nP_n\right)^2. \tag*{(76)} Var ( n ) = ⟨( n − ⟨ n ⟩ ) 2 ⟩ = n ∑ n 2 P n − ( n ∑ n P n ) 2 . (76)
It is the squared rms photon-number fluctuation. For a nonzero mean photon number, F = Var ( n ) / ⟨ n ⟩ F = \operatorname{Var}(n)/\langle n\rangle F = Var ( n ) / ⟨ n ⟩ and g ( 2 ) ( 0 ) = ⟨ n ( n − 1 ) ⟩ / ⟨ n ⟩ 2 g^{(2)}(0) = \langle n(n-1)\rangle/\langle n\rangle^2 g ( 2 ) ( 0 ) = ⟨ n ( n − 1 )⟩ / ⟨ n ⟩ 2 . The coherent ladder-amplitude structure does not prove that the physical reduced field is a coherent state: the joint state Eq. (43) still ties each photon number to a distinct electron recoil. The prescribed-source state and the joint sharp-momentum state are different preparations. For a finite electron packet and vacuum input, the conditional Poisson laws must instead be integrated over the initial momentum density. Explicitly, ⟨ μ ⟩ f 0 = ∫ d p f 0 ( p ) μ ( p ) \langle\mu\rangle_{f_0} = \int dp\,f_0(p)\mu(p) ⟨ μ ⟩ f 0 = ∫ d p f 0 ( p ) μ ( p ) and Var f 0 ( μ ) = ∫ d p f 0 ( p ) [ μ ( p ) − ⟨ μ ⟩ f 0 ] 2 \operatorname{Var}_{f_0}(\mu) = \int dp\,f_0(p)[\mu(p)-\langle\mu\rangle_{f_0}]^2 Var f 0 ( μ ) = ∫ d p f 0 ( p ) [ μ ( p ) − ⟨ μ ⟩ f 0 ] 2 . To derive the variance, use the conditional Poisson identities ⟨ n ⟩ p = μ ( p ) \langle n\rangle_p = \mu(p) ⟨ n ⟩ p = μ ( p ) and ⟨ n 2 ⟩ p = μ ( p ) 2 + μ ( p ) \langle n^2\rangle_p = \mu(p)^2+\mu(p) ⟨ n 2 ⟩ p = μ ( p ) 2 + μ ( p ) . Averaging gives ⟨ n ⟩ = ⟨ μ ⟩ f 0 \langle n\rangle= \langle\mu\rangle_{f_0} ⟨ n ⟩ = ⟨ μ ⟩ f 0 and ⟨ n 2 ⟩ = ⟨ μ 2 ⟩ f 0 + ⟨ μ ⟩ f 0 \langle n^2\rangle= \langle\mu^2\rangle_{f_0}+\langle\mu\rangle_{f_0} ⟨ n 2 ⟩ = ⟨ μ 2 ⟩ f 0 + ⟨ μ ⟩ f 0 . Subtracting the square of the mean yields the law of total variance:
Var ( n ) = ⟨ Var ( n ∣ p ) ⟩ f 0 + Var f 0 ( ⟨ n ∣ p ⟩ ) . (77) \operatorname{Var}(n) = \langle\operatorname{Var}(n|p)\rangle_{f_0} + \operatorname{Var}_{f_0}(\langle n|p\rangle). \tag*{(77)} Var ( n ) = ⟨ Var ( n ∣ p ) ⟩ f 0 + Var f 0 (⟨ n ∣ p ⟩) . (77)
Consequently the complete zeroth-order photon variance is
Var ( n ) = ⟨ μ ⟩ f 0 + Var f 0 ( μ ) . (78) \operatorname{Var}(n) = \langle\mu\rangle_{f_0} + \operatorname{Var}_{f_0}(\mu). \tag*{(78)} Var ( n ) = ⟨ μ ⟩ f 0 + Var f 0 ( μ ) . (78)
For the recoil-corrected vacuum probabilities, write M r = ∑ ℓ ≥ 0 ℓ r P ℓ M_r = \sum_{\ell\ge0}\ell^rP_\ell M r = ∑ ℓ ≥ 0 ℓ r P ℓ ; then Var ( n ) = M 2 − M 1 2 \operatorname{Var}(n) = M_2 - M_1^2 Var ( n ) = M 2 − M 1 2 . If M r = ∑ j = 0 2 ε j M r ( j ) M_r = \sum_{j=0}^2\varepsilon^jM_r^{(j)} M r = ∑ j = 0 2 ε j M r ( j ) , its second-order consistent expression is
Var ( n ) = M 2 ( 0 ) − ( M 1 ( 0 ) ) 2 + ε [ M 2 ( 1 ) − 2 M 1 ( 0 ) M 1 ( 1 ) ] + ε 2 [ M 2 ( 2 ) − 2 M 1 ( 0 ) M 1 ( 2 ) − ( M 1 ( 1 ) ) 2 ] + O ( ε 3 ) . (79) \operatorname{Var}(n) = M_{2}^{(0)} - \left(M_{1}^{(0)}\right)^2 + \varepsilon\left[M_{2}^{(1)} - 2M_{1}^{(0)}M_{1}^{(1)}\right] + \varepsilon^2\left[M_{2}^{(2)} - 2M_{1}^{(0)}M_{1}^{(2)} - \left(M_{1}^{(1)}\right)^2\right] + O(\varepsilon^3). \tag*{(79)} Var ( n ) = M 2 ( 0 ) − ( M 1 ( 0 ) ) 2 + ε [ M 2 ( 1 ) − 2 M 1 ( 0 ) M 1 ( 1 ) ] + ε 2 [ M 2 ( 2 ) − 2 M 1 ( 0 ) M 1 ( 2 ) − ( M 1 ( 1 ) ) 2 ] + O ( ε 3 ) . (79)
A mixture with momentum-dependent μ \mu μ is generally not Poissonian. Its Fano factor is
F = 1 + Var f 0 ( μ ) ⟨ μ ⟩ f 0 ≥ 1. (80) F = 1 + \frac{\operatorname{Var}_{f_0}(\mu)}{\langle\mu\rangle_{f_0}} \ge1. \tag*{(80)} F = 1 + ⟨ μ ⟩ f 0 Var f 0 ( μ ) ≥ 1. (80)
Thus an initial spread can alter photon statistics even when the conditional recoil parameter is zero.
The exact RN Hamiltonian is Hermitian and its evolution unitary. A finite recoil expansion constructed by the Duhamel/Dyson iteration [35 , 36 ] is not exactly unitary: normalization holds only through the retained order, and the residual must be monitored together with boundary population and positivity. Renormalizing a truncated probability distribution does not restore omitted physical terms. The numerical and order-by-order checks reported above apply to the stated parameter range.
Recoil deforms the distribution through the index-dependent coupling phases, not through the removed diagonal gauge phase itself. The expansion is not restricted by definition to small signal. Its control condition is Eq. (51) on every appreciably occupied transition throughout the interval. We make “appreciably occupied” quantitative by choosing a discarded-probability tolerance δ t a i l \delta_{\mathrm{tail}} δ tail and defining the Poisson cutoff
L δ t a i l ( μ ) = min { L ∈ N 0 : ∑ ℓ = 0 L e − μ μ ℓ ℓ ! ≥ 1 − δ t a i l } . (81) L_{\delta_{\mathrm{tail}}}(\mu) = \min\left\{L \in\mathbb{N}_0 : \sum_{\ell=0}^{L} \frac{e^{-\mu}\mu^\ell}{\ell!} \ge1-\delta_{\mathrm{tail}}\right\}. \tag*{(81)} L δ tail ( μ ) = min { L ∈ N 0 : ℓ = 0 ∑ L ℓ ! e − μ μ ℓ ≥ 1 − δ tail } . (81)
For example, δ t a i l = 10 − 6 \delta_{\mathrm{tail}} = 10^{-6} δ tail = 1 0 − 6 gives L δ t a i l = 4 , 9 , 28 , 151 L_{\delta_{\mathrm{tail}}} = 4, 9, 28, 151 L δ tail = 4 , 9 , 28 , 151 for μ = 0.1 , 1 , 10 , 100 \mu= 0.1, 1, 10, 100 μ = 0.1 , 1 , 10 , 100 , respectively. The mean relevant to a finite interval is μ max = max 0 ≤ s ≤ τ max g 2 S 0 ( W , s ) \mu_{\max} = \max_{0 \le s \le\tau_{\max}} g^2 S_0(W,s) μ m a x = max 0 ≤ s ≤ τ m a x g 2 S 0 ( W , s ) , not necessarily its final value. A conservative phase criterion for the retained ladder and its upper neighboring transition is
∣ ε ∣ [ 2 L δ t a i l ( μ max ) + 1 ] τ max ≪ 1. (82) |\varepsilon|\left[2L_{\delta_{\mathrm{tail}}}(\mu_{\max})+1\right]\tau_{\max} \ll1. \tag*{(82)} ∣ ε ∣ [ 2 L δ tail ( μ m a x ) + 1 ] τ m a x ≪ 1. (82)
This replaces an unspecified tail-selection constant by a stated probability tolerance. It is a control estimate based on the recoilless distribution; an actual calculation must also monitor recoil-induced population outside the chosen cutoff and convergence when the cutoff is increased. A norm or observable error test, such as Eq. (66) , gives a separate quantitative accuracy check. For an initial packet the criterion must hold over the retained range of initial momenta as well.
Small signal often keeps the populated ladder short, but is not the mathematical definition of the recoil expansion. In stronger interaction, even a small ∣ ε ∣ |\varepsilon| ∣ ε ∣ can accumulate a significant phase as populated orders move and broaden. In that case retain the factors e i W τ − i ε ( 2 ℓ + 1 ) τ e^{iW\tau-i\varepsilon(2\ell+1)\tau} e iW τ − i ε ( 2 ℓ + 1 ) τ and e − i W τ + i ε ( 2 ℓ − 1 ) τ e^{-iW\tau+i\varepsilon(2\ell-1)\tau} e − iW τ + i ε ( 2 ℓ − 1 ) τ in Eq. (49) without expanding them. These are the complete transition phases: each combines ordinary detuning with the kinetic-energy difference between its neighboring recoil states. Alternatively solve Eq. (47) directly, with a converged ladder cutoff.
Before using a Wigner distribution, it is useful to state what changes when the initial electron is a wave packet. For vacuum input, write the physical state in Eq. (45) as
∣ Ψ ( t ) ⟩ = ∑ ℓ = 0 ∞ ∫ d P χ ℓ ( P , t ) ∣ P , ℓ ⟩ , χ ℓ ( P , t ) = ϕ ( P + ℓ q ) A ℓ ( 0 ) ( P + ℓ q , t ) . (83) |\Psi(t)\rangle= \sum_{\ell=0}^{\infty}\int dP\,\chi_{\ell}(P,t)|P,\ell\rangle,\qquad\chi_{\ell}(P,t)=\phi(P+\ell q)A_{\ell}^{(0)}(P+\ell q,t). \tag*{(83)} ∣Ψ ( t )⟩ = ℓ = 0 ∑ ∞ ∫ d P χ ℓ ( P , t ) ∣ P , ℓ ⟩ , χ ℓ ( P , t ) = ϕ ( P + ℓ q ) A ℓ ( 0 ) ( P + ℓ q , t ) . (83)
A branch labelled ℓ \ell ℓ is the component with exactly ℓ \ell ℓ photons emitted into the initially empty selected mode. Its final electron momentum P P P is the initial momentum p = P + ℓ q p=P+\ell q p = P + ℓ q . Here τ \tau τ is physical time and τ = τ / T 0 \tau=\tau/T_0 τ = τ / T 0 its dimensionless counterpart; C C C takes τ \tau τ as its argument, while χ \chi χ includes the physical free phase and is written as a function of t t t . They describe the same instant. The notation ρ ℓ m ( r ) = ⟨ ℓ ∣ ρ r ∣ m ⟩ \rho_{\ell m}^{(r)}=\langle\ell|\rho_r|m\rangle ρ ℓ m ( r ) = ⟨ ℓ ∣ ρ r ∣ m ⟩ denotes a radiation density-matrix element; r r r labels radiation and ℓ , m \ell,m ℓ , m its row and column photon numbers. Tracing over the appropriate factor gives
ρ m ( r ) ( t ) = ∫ d P χ ℓ ( P , t ) χ m ∗ ( P , t ) , \rho_{m}^{(r)}(t)=\int dP\,\chi_{\ell}(P,t)\chi_{m}^{*}(P,t), ρ m ( r ) ( t ) = ∫ d P χ ℓ ( P , t ) χ m ∗ ( P , t ) ,
ρ e ( P , P ′ , t ) = ∑ ℓ χ ℓ ( P , t ) χ ℓ ∗ ( P ′ , t ) , \rho_e(P,P',t)=\sum_{\ell}\chi_{\ell}(P,t)\chi_{\ell}^{*}(P',t), ρ e ( P , P ′ , t ) = ℓ ∑ χ ℓ ( P , t ) χ ℓ ∗ ( P ′ , t ) ,
P ℓ ( t ) = ∫ d p ∣ ϕ ( p ) ∣ 2 ∣ C ℓ ( 0 ) ( p , t / T 0 ) ∣ 2 . (84) P_{\ell}(t)=\int dp\,|\phi(p)|^2|C_{\ell}^{(0)}(p,t/T_0)|^2. \tag*{(84)} P ℓ ( t ) = ∫ d p ∣ ϕ ( p ) ∣ 2 ∣ C ℓ ( 0 ) ( p , t / T 0 ) ∣ 2 . (84)
These formulas separate the roles of the initial state. Photon-number probabilities for vacuum input average the conditional probabilities over f 0 ( p ) = ∣ ϕ ( p ) ∣ 2 f_0(p)=|\phi(p)|^2 f 0 ( p ) = ∣ ϕ ( p ) ∣ 2 , whereas optical off-diagonal elements depend on overlap of the recoil-shifted electron packets. A coherent packet and an incoherent momentum mixture with the same f 0 f_0 f 0 therefore give the same unconditional photon-number probabilities in this translation-invariant vacuum-input model, but need not give the same optical coherence or electron spatial structure. This equivalence does not extend automatically to a coherent optical seed, where different initial photon sectors can interfere.
For illustration, a minimum-uncertainty Gaussian electron packet has
ϕ ( p ) = 1 ( 2 π σ p 0 2 ) 1 / 4 exp [ − ( p − p c ) 2 4 σ p 0 2 − i p z c ℏ ] , σ z 0 = ℏ 2 σ p 0 . (85) \phi(p)=\frac{1}{(2\pi\sigma_{p0}^{2})^{1/4}}\exp\left[-\frac{(p-p_c)^2}{4\sigma_{p0}^{2}}-\frac{ipz_c}{\hbar}\right],\qquad\sigma_{z0}=\frac{\hbar}{2\sigma_{p0}}. \tag*{(85)} ϕ ( p ) = ( 2 π σ p 0 2 ) 1/4 1 exp [ − 4 σ p 0 2 ( p − p c ) 2 − ℏ i p z c ] , σ z 0 = 2 σ p 0 ℏ . (85)
If the conditional physical amplitudes vary negligibly across the relevant shifted packet supports, the overlap factor multiplying A ℓ A m ∗ A_{\ell}A_{m}^{*} A ℓ A m ∗ is
∫ d P ϕ ( P + ℓ q ) ϕ ∗ ( P + m q ) = exp [ − ( ℓ − m ) 2 q 2 8 σ p 0 2 − i ( ℓ − m ) q z c ℏ ] . (86) \int dP\,\phi(P+\ell q)\phi^{*}(P+mq)=\exp\left[-\frac{(\ell-m)^2q^2}{8\sigma_{p0}^{2}}-\frac{i(\ell-m)qz_c}{\hbar}\right]. \tag*{(86)} ∫ d P ϕ ( P + ℓ q ) ϕ ∗ ( P + m q ) = exp [ − 8 σ p 0 2 ( ℓ − m ) 2 q 2 − ℏ i ( ℓ − m ) q z c ] . (86)
For σ p 0 ≪ q \sigma_{p0}\ll q σ p 0 ≪ q , different recoil orders leave nearly distinguishable electron states and optical coherences are suppressed. For broader packets their overlap is larger, but the spread of detunings may then spoil the assumption of constant amplitudes. The full integrals in Eq. (84) , including the momentum-dependent free phase, must be used whenever this approximation fails.
Initial uncertainty and interaction-generated fluctuations must consequently be kept distinct. The Gaussian has an initial momentum variance σ p 0 2 \sigma_{p0}^{2} σ p 0 2 and a corresponding position uncertainty before any FEL emission. Recoil subsequently changes its momentum populations and correlations. For a phase-defined optical input, initial electron coherence can also contribute immediately to the photon change:
d ⟨ n ^ r ⟩ d t ∣ t = 0 = 2 g Im [ α 0 ∗ ⟨ T ^ q ⟩ 0 ] . (87) \left.\frac{d\langle\hat n_r\rangle}{dt}\right|_{t=0}=2g\,\operatorname{Im}\left[\alpha_0^{*}\langle\hat T_q\rangle_0\right]. \tag*{(87)} d t d ⟨ n ^ r ⟩ t = 0 = 2 g Im [ α 0 ∗ ⟨ T ^ q ⟩ 0 ] . (87)
Explicitly, since [ a ^ † , n ^ r ] = − a ^ † [\hat a^{\dagger},\hat n_r]=-\hat a^{\dagger} [ a ^ † , n ^ r ] = − a ^ † and [ a ^ , n ^ r ] = a ^ [\hat a,\hat n_r]=\hat a [ a ^ , n ^ r ] = a ^ , the Heisenberg equation gives
d ⟨ n ^ r ⟩ d t = i g ⟨ − a ^ † T ^ q + a ^ T ^ q † ⟩ . \frac{d\langle\hat n_r\rangle}{dt}=ig\langle-\hat a^{\dagger}\hat T_q+\hat a\hat T_q^{\dagger}\rangle. d t d ⟨ n ^ r ⟩ = i g ⟨ − a ^ † T ^ q + a ^ T ^ q † ⟩ .
For the initial product state the expectation factors, with ⟨ a ^ ⟩ = α 0 \langle\hat a\rangle=\alpha_0 ⟨ a ^ ⟩ = α 0 and ⟨ T ^ q † ⟩ = ⟨ T ^ q ⟩ ∗ \langle\hat T_q^{\dagger}\rangle=\langle\hat T_q\rangle^{*} ⟨ T ^ q † ⟩ = ⟨ T ^ q ⟩ ∗ . Substitution yields Eq. (87) . It vanishes for vacuum input or a momentum-diagonal electron state. Thus the weak-transition number-state expressions Eq. (64) and Eq. (65) cannot simply be reused for an arbitrary phase-localized electron and coherent seed by replacing n n n with ∣ α 0 ∣ 2 |\alpha_0|^2 ∣ α 0 ∣ 2 .
A second limitation is that the interaction itself changes the electron momentum distribution. Define
σ p 2 ( t ) = ⟨ p 2 ⟩ − ⟨ p ⟩ 2 , Δ ( p ) = E ( p − q ) − E ( p ) ℏ + ω . (88) \sigma_p^2(t)=\langle p^2\rangle-\langle p\rangle^2,\qquad\Delta(p)=\frac{E(p-q)-E(p)}{\hbar}+\omega. \tag*{(88)} σ p 2 ( t ) = ⟨ p 2 ⟩ − ⟨ p ⟩ 2 , Δ ( p ) = ℏ E ( p − q ) − E ( p ) + ω . (88)
For vacuum input, the joint outcome distribution of initial momentum and exchanged photon number is ∣ ϕ ( p ) ∣ 2 ∣ C ℓ ( 0 ) ( p , t / T 0 ) ∣ 2 |\phi(p)|^2|C_{\ell}^{(0)}(p,t/T_0)|^2 ∣ ϕ ( p ) ∣ 2 ∣ C ℓ ( 0 ) ( p , t / T 0 ) ∣ 2 . Since the outgoing momentum is P = p − q ℓ P=p-q\ell P = p − q ℓ , its variance satisfies
σ p 2 ( t ) = σ p 0 2 + q 2 Var ( ℓ ) − 2 q Cov ( p , ℓ ) . (89) \sigma_p^2(t)=\sigma_{p0}^2+q^2\operatorname{Var}(\ell)-2q\operatorname{Cov}(p,\ell). \tag*{(89)} σ p 2 ( t ) = σ p 0 2 + q 2 Var ( ℓ ) − 2 q Cov ( p , ℓ ) . (89)
The covariance is evaluated in that joint outcome distribution. If the exchange probabilities are independent of the initial momentum, it vanishes and the recoil contribution adds to the initial variance. If detuning selects parts of the packet, the covariance must be retained; a broadening formula based only on q 2 Var ( ℓ ) q^2\operatorname{Var}(\ell) q 2 Var ( ℓ ) is then incomplete.
Near a reference momentum p r p_r p r ,
Δ ( p ) ≃ Δ ( p r ) + Δ ′ ( p r ) ( p − p r ) , σ Δ ( t ) ≃ ∣ Δ ′ ( p r ) ∣ σ p ( t ) . (90) \Delta(p)\simeq\Delta(p_r)+\Delta'(p_r)(p-p_r),\qquad\sigma_\Delta(t)\simeq\lvert\Delta'(p_r)\rvert\sigma_p(t). \tag*{(90)} Δ ( p ) ≃ Δ ( p r ) + Δ ′ ( p r ) ( p − p r ) , σ Δ ( t ) ≃ ∣ Δ ′ ( p r )∣ σ p ( t ) . (90)
The two dispersions are physically distinct: recoil changes the neighboring ladder mismatch, while electron momentum spread distributes the mismatch across the beam. Both can alter observable quantum signatures. The initial width exists before radiation is exchanged, while the kinetic recoil correction changes the energy spacing of successive exchange states. They can produce similar changes in a measured spectrum, but have different origins and cannot be represented by the same parameter.
For an externally specified initial distribution one can average conditional probabilities,
P ℓ ‾ ( t ) = ∫ d p f 0 ( p ) P ℓ ( t ∣ p ) . (91) \overline{P_\ell}(t)=\int dp\,f_0(p)P_\ell(t\mid p). \tag*{(91)} P ℓ ( t ) = ∫ d p f 0 ( p ) P ℓ ( t ∣ p ) . (91)
For vacuum input this is exactly the photon-number marginal in Eq. (84) , for either a pure packet or a momentum mixture with diagonal density f 0 f_0 f 0 . Phase-sensitive observables require the density-matrix information as well. This averaging does not imply that the outgoing electrons remain unchanged. Low-gain response already produces perturbative energy modulation and distribution changes. The present calculation describes one quantum electron, or an independent statistical ensemble of such single-electron passages in a prescribed field. It does not include collective gain. A many-electron field evolving self-consistently with electron correlations is a separate extension, and cannot be inferred from this single-electron average.
The Gaussian packet also makes the initial phase mixing explicit. Linearizing W ( p ) W(p) W ( p ) about p c p_c p c , define W c = W ( p c ) W_c=W(p_c) W c = W ( p c ) and σ W = ∣ ∂ p W ( p c ) ∣ σ p 0 \sigma_W=\lvert\partial_pW(p_c)\rvert\sigma_{p0} σ W = ∣ ∂ p W ( p c )∣ σ p 0 . The weak-transition recoilless line shape averaged over this packet is
⟨ S 0 ( W ( p ) , τ ) ⟩ f 0 = ∫ 0 τ d u ∫ 0 τ d v exp [ i W c ( u − v ) − σ W 2 ( u − v ) 2 2 ] . (92) \langle S_0(W(p),\tau)\rangle_{f_0}
=\int_0^\tau du\int_0^\tau dv\,\exp\left[iW_c(u-v)-\frac{\sigma_W^2(u-v)^2}{2}\right]. \tag*{(92)} ⟨ S 0 ( W ( p ) , τ ) ⟩ f 0 = ∫ 0 τ d u ∫ 0 τ d v exp [ i W c ( u − v ) − 2 σ W 2 ( u − v ) 2 ] . (92)
The quantity averaged here is the conditional finite-time spectral factor S 0 ( W , τ ) = τ 2 sinc 2 ( W τ / 2 ) S_0(W,\tau)=\tau^2\operatorname{sinc}^2(W\tau/2) S 0 ( W , τ ) = τ 2 sinc 2 ( W τ /2 ) of Eq. (59) , rather than a radiation-field amplitude averaged over distinct electrons. For vacuum input, weak coupling and zero recoil, the selected-mode spontaneous probability is g 2 ⟨ S 0 ⟩ f 0 g^2\langle S_0\rangle_{f_0} g 2 ⟨ S 0 ⟩ f 0 to leading order. Thus the answer is indeed the spontaneous-emission line shape convolved with the entrance momentum density. For a fixed initial photon number n n n , the same zero-recoil kernel multiplies ( n + 1 ) g 2 (n+1)g^2 ( n + 1 ) g 2 for emission and n g 2 ng^2 n g 2 for absorption. It therefore also appears in stimulated transitions; the interpretation applies purely to vacuum input. Recoil separates the emission and absorption detunings as established earlier.
This follows by performing the Gaussian momentum average of the two-time phase factor. The factor involving σ W \sigma_W σ W suppresses contributions separated by more than 1 / σ W 1/\sigma_W 1/ σ W in the dimensionless time τ = t / T 0 \tau=t/T_0 τ = t / T 0 ; the corresponding physical separation is T 0 / σ W T_0/\sigma_W T 0 / σ W . With σ W = 0 \sigma_W=0 σ W = 0 , the condition line shape is recovered. With recoil retained and nearly constant across the packet, the elementary emission or absorption kernel uses the corresponding shifted central detuning. Thus initial momentum uncertainty changes the finite-time spectrum as well as the visibility of individual recoil orders.
Changing variables to s = u − v s=u-v s = u − v gives a convenient real integral for the numerical comparison,
⟨ S 0 ⟩ f 0 = 2 ∫ 0 τ ( τ − s ) cos ( W c s ) e − σ W 2 s 2 / 2 d s . (93) \langle S_0\rangle_{f_0}=2\int_0^\tau(\tau-s)\cos(W_cs)e^{-\sigma_W^2s^2/2}\,ds. \tag*{(93)} ⟨ S 0 ⟩ f 0 = 2 ∫ 0 τ ( τ − s ) cos ( W c s ) e − σ W 2 s 2 /2 d s . (93)
Figure 1 compares this expression at τ = 1 \tau=1 τ = 1 with the unaveraged kernel for several entrance widths. Momentum averaging lowers the central peak and smooths the zeros and side lobes; it broadens the spectrum without increasing its integrated weight. Indeed, ∫ d W c ⟨ S 0 ⟩ = 2 π \int dW_c\langle S_0\rangle=2\pi ∫ d W c ⟨ S 0 ⟩ = 2 π for every normalized Gaussian convolution. A change of peak height or spectral width in this calculation is consequently an entrance-distribution effect and does not by itself identify quantum recoil.
Figure 1. Conditional and Gaussian-averaged zero-recoil line shapes at τ = 1 \tau= 1 τ = 1 . The solid reference is S 0 ( W c , 1 ) = sinc 2 ( W c / 2 ) S_0(W_c,1) = \operatorname{sinc}^2(W_c/2) S 0 ( W c , 1 ) = sinc 2 ( W c /2 ) ; the other curves use Eq. (93) with dimensionless rms detuning widths σ W = 0.5 , 1 , 2 , 4 \sigma_W = 0.5, 1, 2, 4 σ W = 0.5 , 1 , 2 , 4 . The ordinate is the spectral factor, without the coupling or photon-number multiplier.
The initial-state question has an operational consequence. Conventional beam measurements of energy spread, emittance and bunch duration generally do not reconstruct the complete single-electron density operator. In particular, the ensemble momentum spread need not equal the intrinsic momentum uncertainty of each electron packet: it can include variation of packet centers from electron to electron. Choosing every electron as a minimum-uncertainty Gaussian, or every electron as a sharp momentum state, is therefore a model assumption unless the preparation or additional measurements support it. The uncertainty relation bounds compatible variances but does not select a pure state or impose equality. Nor is a decomposition of a mixed density operator into particular packets uniquely physical; all predictions depend on the density operator itself.
This incomplete specification does not make the recoil Hamiltonian or quantum corrections intrinsically ambiguous. It makes some predictions conditional on information about the preparation. The distinction is especially clear in the vacuum-input, translation-invariant problem already solved. For any normalized initial electron density operator, its photon-number probabilities are
P ℓ ( t ) = ∫ d p ρ e 0 ( p , p ) ∣ C ℓ ( 0 ) ( p , t / T 0 ) ∣ 2 . (94) P_\ell(t) = \int dp\,\rho_{e0}(p,p)\left|\mathcal{C}^{(0)}_\ell(p,t/T_0)\right|^2. \tag*{(94)} P ℓ ( t ) = ∫ d p ρ e 0 ( p , p ) C ℓ ( 0 ) ( p , t / T 0 ) 2 . (94)
The momentum diagonal is understood as a probability density. Pure packets and mixed preparations with the same diagonal give the same counts, even when their position coherence is different. The conditional recoil-dependent line shape and its average are consequently meaningful without assuming a minimum-uncertainty packet. This statement uses vacuum input, the specified interaction and unconditional photon counting; it is not a state-independence theorem for all FEL observables.
Optical phase, electron bunching and coherently seeded response require more information. For example,
⟨ T ^ q ⟩ 0 = ∫ d p ρ e 0 ( p , p − q ) (95) \langle\hat{T}_q\rangle_0 = \int dp\,\rho_{e0}(p,p-q) \tag*{(95)} ⟨ T ^ q ⟩ 0 = ∫ d p ρ e 0 ( p , p − q ) (95)
contains electron momentum coherence at separation q q q . It is not fixed by the measured momentum diagonal. Randomizing the spatial centers of localized packets can leave that diagonal unchanged while suppressing Eq. (95) ; Eq. (87) then gives a different phase-sensitive response. Our Gaussian and sharp-momentum examples illustrate these distinct possibilities rather than competing claims about an unmeasured beam state. A fully delocalized ideal momentum state must also not be confused with the measured finite duration of an ensemble bunch.
A useful way to report this preparation uncertainty is to define the set S S S of positive, normalized initial density operators compatible with the measured beam information. For an observable O O O , its predicted range is
O min = inf ρ e 0 ∈ S O [ ρ e 0 ] , O max = sup ρ e 0 ∈ S O [ ρ e 0 ] . (96) O_{\min}=\inf_{\rho_{e0}\in S}O[\rho_{e0}],\qquad O_{\max}=\sup_{\rho_{e0}\in S}O[\rho_{e0}]. \tag*{(96)} O m i n = ρ e 0 ∈ S inf O [ ρ e 0 ] , O m a x = ρ e 0 ∈ S sup O [ ρ e 0 ] . (96)
A classical comparison uses a statistical ensemble constrained by the entrance measurements, rather than a classical object representing an arbitrary pure quantum state. For the minimum-uncertainty Gaussian used here, the initial Wigner function is positive and can be taken directly as a normalized classical phase-space density with the same position and momentum marginals. The quantum prediction evolves it with the joint Hamiltonian or the appropriate Wigner equation; the classical prediction evolves the same density with Liouville transport and, for collective gain, the same Maxwell–beam feedback. Both calculations use the same seed, cavity losses, mode selection and detector response. In a semiclassical photon-count baseline, a conditional coherent intensity μ \mu μ yields Poisson counts; its classical fluctuations must also be averaged.
For a non-Gaussian quantum preparation with a negative Wigner function there is no identical positive classical phase-space density. One instead compares the measured observable with the family of positive classical distributions compatible with the independently measured entrance data. Such a comparison is operational, and does not assume that every quantum preparation admits a classical representation. The particular baseline must be stated: otherwise a difference may be due to unequal entrance information rather than the recoil dynamics being tested. A proposed recoil signature should therefore be compared with classical predictions using compatible initial beam data and the same propagation and detection model. If the ranges overlap within experimental uncertainty, that observable does not establish the proposed quantum correction under the available characterization. One must either constrain the missing coherences or use an observable insensitive to them. Recoil-dependent unconditional counts in Eq. (94) provide an example of the latter within the model of one electron traversing the interaction region once with the selected radiation mode initially in vacuum. A reliably measured F < 1 F<1 F < 1 is a different kind of evidence: it excludes a classical statistical mixture of coherent radiation states, but does not by itself identify the initial electron state or establish that a particular recoil mechanism caused it. Unknown preparation thus limits inference for specified observables; it does not invalidate the quantum description.
A dimensionless dephasing measure is
D ( t ) = σ Δ ( t ) T c o h , (97) \mathcal{D}(t)=\sigma_{\Delta}(t)T_{\mathrm{coh}}, \tag*{(97)} D ( t ) = σ Δ ( t ) T coh , (97)
where T c o h T_{\mathrm{coh}} T coh is the physical interaction or coherence time in seconds. With normalized frequencies one equivalently uses σ Δ 0 ( t ) ( T c o h / T 0 ) \sigma_{\Delta_0}(t)(T_{\mathrm{coh}}/T_0) σ Δ 0 ( t ) ( T coh / T 0 ) . Here σ Δ \sigma_{\Delta} σ Δ is an angular-frequency spread in s − 1 \mathrm{s}^{-1} s − 1 , so its product with a time in seconds is dimensionless. If an energy-mismatch spread σ δ E \sigma_{\delta E} σ δ E is used instead, the same measure is σ δ E T c o h / ℏ \sigma_{\delta E}T_{\mathrm{coh}}/\hbar σ δ E T coh /ℏ . For a static Gaussian mismatch distribution, ⟨ e i Δ t ⟩ = e i ⟨ Δ ⟩ t − σ Δ 2 t 2 / 2 \langle e^{i\Delta t}\rangle=e^{i\langle\Delta\rangle t-\sigma_{\Delta}^{2}t^{2}/2} ⟨ e i Δ t ⟩ = e i ⟨ Δ ⟩ t − σ Δ 2 t 2 /2 , so D ≪ 1 \mathcal{D}\ll1 D ≪ 1 implies weak phase dispersion over the coherence interval, whereas D ≳ 1 \mathcal{D}\gtrsim1 D ≳ 1 implies appreciable phase mixing. (92) derives this estimate explicitly in the present model. The width must be followed during the interaction: a narrow entrance packet does not guarantee a narrow distribution after stimulated modulation or spontaneous diffusion.
Wigner phase space, modulation and bunching
The Wigner representation provides a bridge between the recoil ladder and classical phase-space transport [40 , 41 , 42 ] . To avoid treating a dimensionless phase as a dimensional canonical coordinate, we first use the pair ( z , p ) (z,p) ( z , p ) with [ z ^ , p ^ ] = i ℏ [\hat{z},\hat{p}]=i\hbar [ z ^ , p ^ ] = i ℏ and define
W ( z , p , t ) = 1 2 π ℏ ∫ d ξ e − i p ξ / ℏ ⟨ z + ξ 2 | ρ ^ e ( t ) | z − ξ 2 ⟩ . (98) W(z,p,t)=\frac{1}{2\pi\hbar}\int d\xi\,e^{-ip\xi/\hbar}\left\langle z+\frac{\xi}{2}\middle|\hat{\rho}_{e}(t)\middle|z-\frac{\xi}{2}\right\rangle. \tag*{(98)} W ( z , p , t ) = 2 π ℏ 1 ∫ d ξ e − i p ξ /ℏ ⟨ z + 2 ξ ρ ^ e ( t ) z − 2 ξ ⟩ . (98)
For the vacuum-input packet already propagated in Section VI, its momentum-space form follows directly from Eq. (84) :
W ( z , p , t ) = 1 2 π ℏ ∑ ℓ ∫ d k e i k z / ℏ χ ℓ ( p + κ / 2 , t ) χ ℓ ∗ ( p − κ / 2 , t ) . (99) W(z,p,t)=\frac{1}{2\pi\hbar}\sum_{\ell}\int dk\,e^{ikz/\hbar}\chi_{\ell}(p+\kappa/2,t)\chi_{\ell}^{*}(p-\kappa/2,t). \tag*{(99)} W ( z , p , t ) = 2 π ℏ 1 ℓ ∑ ∫ d k e ik z /ℏ χ ℓ ( p + κ /2 , t ) χ ℓ ∗ ( p − κ /2 , t ) . (99)
The variable κ \kappa κ is a momentum difference. This is the Weyl representation of the electron density matrix obtained from the Raman–Nath solution, not an independently postulated initial distribution. For the Gaussian Eq. (85) , its initial value is
W 0 ( z , p ) = 1 π ℏ exp [ − ( p − p c ) 2 2 σ p 0 2 − 2 σ p 0 2 ( z − z c ) 2 ℏ 2 ] . (100) W_0(z,p)=\frac{1}{\pi\hbar}\exp\left[-\frac{(p-p_c)^2}{2\sigma_{p0}^2}-\frac{2\sigma_{p0}^2(z-z_c)^2}{\hbar^2}\right]. \tag*{(100)} W 0 ( z , p ) = π ℏ 1 exp [ − 2 σ p 0 2 ( p − p c ) 2 − ℏ 2 2 σ p 0 2 ( z − z c ) 2 ] . (100)
At later times Eq. (99) retains the recoil-branch weights, packet spreading and momentum coherences supplied by the conditional amplitudes. A semiclassical closed Wigner equation, discussed below, is a further approximation to this fully joint construction.
Its marginals are the position and momentum probability densities. It is normalized for a normalized electron state but need not be positive everywhere; it is a quasiprobability. The phase θ = q z / ℏ \theta=qz/\hbar θ = q z /ℏ can subsequently be introduced with a corresponding canonical momentum rescaling.
For Weyl symbols A A A and B B B , the star product and Moyal bracket are defined by [41 ]
A ⋆ B = A exp [ i ℏ 2 ( ∂ ← z ∂ → p − ∂ ← p ∂ → z ) ] B , (101) A\star B=A\exp\left[\frac{i\hbar}{2}\left(\overleftarrow{\partial}_z\overrightarrow{\partial}_p-\overleftarrow{\partial}_p\overrightarrow{\partial}_z\right)\right]B, \tag*{(101)} A ⋆ B = A exp [ 2 i ℏ ( ∂ z ∂ p − ∂ p ∂ z ) ] B , (101)
{ A , B } M B = A ⋆ B − B ⋆ A i ℏ . (102) \{A,B\}_{\mathrm{MB}}=\frac{A\star B-B\star A}{i\hbar}. \tag*{(102)} { A , B } MB = i ℏ A ⋆ B − B ⋆ A . (102)
The arrows state which factor is differentiated. The operator evolution i ℏ ρ ˙ e = [ H ^ , ρ e ] i\hbar\dot{\rho}_e=[\hat{H},\rho_e] i ℏ ρ ˙ e = [ H ^ , ρ e ] is the von Neumann equation. Its Weyl transform is the Wigner–Moyal equation below. The finite-difference potential term derived subsequently is a contribution to that transformed equation, rather than a separate operator von Neumann equation. The Moyal bracket approaches the Poisson bracket as the higher-order terms become negligible. Applying the Weyl transform to ρ ˙ e = − i [ H ^ , ρ e ] / ℏ \dot{\rho}_e=-i[\hat{H},\rho_e]/\hbar ρ ˙ e = − i [ H ^ , ρ e ] /ℏ gives
∂ t W = { H , W } M B , H ( z , p , t ) = E ( p ) + V ( z , t ) . (103) \partial_t W=\{H,W\}_{\mathrm{MB}},\qquad H(z,p,t)=E(p)+V(z,t). \tag*{(103)} ∂ t W = { H , W } MB , H ( z , p , t ) = E ( p ) + V ( z , t ) . (103)
This exact equation assumes a specified scalar electron Hamiltonian. In an entangled electron–field problem the reduced electron equation need not close on W W W alone. Its coupling contains field-conditioned electron correlations unless a semiclassical or mean-field reduction is made [25 , 33 ] .
The potential coefficient formerly left schematic can be derived explicitly. In a semiclassical field approximation replace a ^ \hat{a} a ^ by a complex amplitude α ( t ) \alpha(t) α ( t ) . The interaction in Eq. (19) becomes
V ( z , t ) = V − ( t ) e − i q z / ℏ + V − ∗ ( t ) e i q z / ℏ , V − ( t ) = ℏ g α ∗ ( t ) . (104) V(z,t)=V_-(t)e^{-iqz/\hbar}+V_-^*(t)e^{iqz/\hbar},\qquad V_-(t)=\hbar g\alpha^*(t). \tag*{(104)} V ( z , t ) = V − ( t ) e − i q z /ℏ + V − ∗ ( t ) e i q z /ℏ , V − ( t ) = ℏ g α ∗ ( t ) . (104)
All quantities are dimensional here. The star-product translation identities are
e − i q z / ℏ ⋆ W = e − i q z / ℏ W ( z , p + q / 2 , t ) , e^{-iqz/\hbar}\star W=e^{-iqz/\hbar}W(z,p+q/2,t), e − i q z /ℏ ⋆ W = e − i q z /ℏ W ( z , p + q /2 , t ) ,
W ⋆ e − i q z / ℏ = e − i q z / ℏ W ( z , p − q / 2 , t ) . (105) W\star e^{-iqz/\hbar}=e^{-iqz/\hbar}W(z,p-q/2,t). \tag*{(105)} W ⋆ e − i q z /ℏ = e − i q z /ℏ W ( z , p − q /2 , t ) . (105)
For the conjugate harmonic the two shifts are interchanged. Substitution into Eq. (102) gives the exact potential contribution
∂ t W ∣ V = V − e − i q z / ℏ − V − ∗ e i q z / ℏ i ℏ [ W ( z , p + q / 2 , t ) − W ( z , p − q / 2 , t ) ] = 2 g Im [ α ∗ ( t ) e − i q z / ℏ ] [ W ( z , p + q / 2 , t ) − W ( z , p − q / 2 , t ) ] . (106) \begin{aligned}
\left.\partial_tW\right|_V=\frac{V_-e^{-iqz/\hbar}-V_-^*e^{iqz/\hbar}}{i\hbar}\left[W(z,p+q/2,t)-W(z,p-q/2,t)\right] \\
&=2g\,\operatorname{Im}\left[\alpha^*(t)e^{-iqz/\hbar}\right]\left[W(z,p+q/2,t)-W(z,p-q/2,t)\right]. \tag*{(106)}
\end{aligned} ∂ t W ∣ V = i ℏ V − e − i q z /ℏ − V − ∗ e i q z /ℏ [ W ( z , p + q /2 , t ) − W ( z , p − q /2 , t ) ] = 2 g Im [ α ∗ ( t ) e − i q z /ℏ ] [ W ( z , p + q /2 , t ) − W ( z , p − q /2 , t ) ] . (106)
Thus the coefficient, sign and normalization follow from the same Hamiltonian. Quantum recoil appears as a finite momentum translation. The shift is q / 2 q/2 q /2 in the Wigner equation because the density matrix combines two momentum arguments; the elementary electron transition still changes momentum by q q q .
For a quadratic kinetic approximation E ( p ) ≃ E 0 + v 0 ( p − p 0 ) + ( p − p 0 ) 2 / ( 2 m ∥ ) E(p) \simeq E_0+v_0(p-p_0)+(p-p_0)^2/(2m_{\parallel}) E ( p ) ≃ E 0 + v 0 ( p − p 0 ) + ( p − p 0 ) 2 / ( 2 m ∥ ) , the free contribution is exactly − [ v 0 + ( p − p 0 ) / m ∥ ] ∂ z W -[v_0+(p-p_0)/m_{\parallel}]\partial_z\mathcal{W} − [ v 0 + ( p − p 0 ) / m ∥ ] ∂ z W . For the full relativistic dispersion the kinetic star commutator must also be retained. Its derivative expansion begins
∂ t W ∣ E = − E ′ ( p ) ∂ z W + ℏ 2 24 E ′ ′ ′ ( p ) ∂ z 3 W + ⋯ (107) \left.\partial_t\mathcal{W}\right|_E=-E'(p)\partial_z\mathcal{W}+\frac{\hbar^2}{24}E'''(p)\partial_z^3\mathcal{W}+\cdots \tag*{(107)} ∂ t W ∣ E = − E ′ ( p ) ∂ z W + 24 ℏ 2 E ′′′ ( p ) ∂ z 3 W + ⋯ (107)
It would be inconsistent to describe the potential term as exact while silently replacing a non-quadratic kinetic term by purely classical drift. Expanding the finite difference in Eq. (106) ,
W ( p + q / 2 ) − W ( p − q / 2 ) = q ∂ p W + q 3 24 ∂ p 3 W + q 5 1920 ∂ p 5 W + ⋯ . (108) \mathcal{W}(p+q/2)-\mathcal{W}(p-q/2)=q\partial_p\mathcal{W}+\frac{q^3}{24}\partial_p^3\mathcal{W}+\frac{q^5}{1920}\partial_p^5\mathcal{W}+\cdots. \tag*{(108)} W ( p + q /2 ) − W ( p − q /2 ) = q ∂ p W + 24 q 3 ∂ p 3 W + 1920 q 5 ∂ p 5 W + ⋯ . (108)
Since ∂ z V = q [ V − e − i q z / ℏ − V ∗ e i q z / ℏ ] / ( i ℏ ) \partial_zV=q[V_-e^{-iqz/\hbar}-V_*e^{iqz/\hbar}]/(i\hbar) ∂ z V = q [ V − e − i q z /ℏ − V ∗ e i q z /ℏ ] / ( i ℏ ) , the leading potential term is ∂ z V ∂ p W \partial_zV\partial_p\mathcal{W} ∂ z V ∂ p W . Together with the leading kinetic term it gives
∂ t W + E ′ ( p ) ∂ z W − ∂ z V ∂ p W = 0 , (109) \partial_t\mathcal{W}+E'(p)\partial_z\mathcal{W}-\partial_zV\partial_p\mathcal{W}=0, \tag*{(109)} ∂ t W + E ′ ( p ) ∂ z W − ∂ z V ∂ p W = 0 , (109)
the classical Liouville/Vlasov transport equation for the specified Hamiltonian.
A closely related low-gain phase-space analysis is given by Carmesin, Kling, Giese, Sauerbrey and Schleich [43 ] . They compare quantum Wigner evolution with classical trajectories in the same periodic potential and show that populating many momentum levels is not sufficient for classical correspondence. In their notation α W = U 0 / ( 2 ℏ ω r ) \alpha_W=U_0/(2\hbar\omega_r) α W = U 0 / ( 2ℏ ω r ) measures potential height relative to recoil energy; it is distinct from the oscillator parameter α n \alpha_n α n of Section XII. A large α W \alpha_W α W must be accompanied by a distribution smooth on the recoil momentum scale. Their derivative expansion is therefore consistent with the finite-difference criterion derived below. They also show that phase-space quantum structure can be suppressed in averaged observables such as gain. This complements our comparison between RN ladder amplitudes, reduced electron distributions and radiation statistics: agreement of a gain curve does not imply agreement of the full quantum state. The importance of this work is therefore both conceptual and experimental: it identifies which phase-space features are removed by beam averaging and motivates measurements beyond an integrated first moment. It is a phase-space analysis, not itself a demonstration of a few-photon cavity experiment.
A complementary three-dimensional discrete-Wigner treatment by Piovella and collaborators includes longitudinal momentum quantization and transverse geometry [44 ] ; it addresses a broader beam-evolution problem than the low-gain single-electron comparison considered here.
Their phase-uniform initial distribution, W ( θ , P ; 0 ) = ρ ( P ) / ( 2 π ) \mathcal{W}(\theta,\mathcal{P};0)=\rho(\mathcal{P})/(2\pi) W ( θ , P ; 0 ) = ρ ( P ) / ( 2 π ) , represents a statistical ensemble; position and momentum widths need not be tied as in a pure minimum-uncertainty packet. Our RN packet calculations retain a specified quantum input and its recoil coherences, whereas their Wigner treatment emphasizes ensemble averaging and uses a semiclassical radiation amplitude. It is consequently a useful comparison for electron phase-space dynamics, not a calculation of the fully quantized cavity photon distribution. The two approaches share the recoil ladder but answer different observable questions.
The higher odd derivatives give quantum corrections. Replacing the potential difference by its first derivative requires smoothness on the recoil scale,
R q = q Δ p c h a r ≪ 1. (110) R_q=\frac{q}{\Delta p_{\mathrm{char}}}\ll1. \tag*{(110)} R q = Δ p char q ≪ 1. (110)
Here Δ p c h a r \Delta p_{\mathrm{char}} Δ p char is the local momentum scale over which the Wigner function varies appreciably in the region contributing to the observable. It is not automatically the rms beam spread: interference fringes or narrow substructure can have much smaller scales. Quantitatively, where derivative ratios are meaningful, ( Δ p c h a r ) − 2 ∼ ∣ ∂ p 3 W / ∂ p W ∣ (\Delta p_{\mathrm{char}})^{-2}\sim|\partial_p^3\mathcal{W}/\partial_p\mathcal{W}| ( Δ p char ) − 2 ∼ ∣ ∂ p 3 W / ∂ p W ∣ gives the leading correction relative to the first derivative. The leading omitted potential term is then of relative order q 2 / ( 24 Δ p c h a r 2 ) q^2/(24\Delta p_{\mathrm{char}}^2) q 2 / ( 24Δ p char 2 ) .
For nonquadratic free dispersion a spatial smoothness condition must also be checked. If L c h a r L_{\mathrm{char}} L char is the spatial scale of variation of W \mathcal{W} W , the leading kinetic correction relative to classical drift is of order
R z ∼ ℏ 2 24 L c h a r 2 ∣ E ( 3 ) ( p ) E ′ ( p ) ∣ ≪ 1. (111) R_z\sim\frac{\hbar^2}{24L_{\mathrm{char}}^2}\left|\frac{E^{(3)}(p)}{E'(p)}\right|\ll1. \tag*{(111)} R z ∼ 24 L char 2 ℏ 2 E ′ ( p ) E ( 3 ) ( p ) ≪ 1. (111)
This estimate applies where the leading drift is nonzero; otherwise the kinetic terms must be compared directly. Higher derivatives require analogous checks. For a strictly quadratic E E E , all higher kinetic corrections vanish, so no additional spatial restriction follows from that part of the Moyal expansion. This criterion concerns spatial structure of the Wigner function relative to the dispersion scale, and does not universally require a packet longer than the optical grating period. This limit is conceptually distinct from the phase expansion in Eq. (51) . The latter controls a finite-time approximation to the Raman–Nath hopping phases; the former controls a phase-space derivative expansion. The neighboring-state energy differences already derived in Eq. (35) and Eq. (50) supply their common recoil scale, so they need not be repeated here.
The first momentum moments and bunching are
⟨ p ⟩ = ∫ d z d p p W , σ p 2 = ∫ d z d p ( p − ⟨ p ⟩ ) 2 W , b − = ∫ d z d p e − i q z / ℏ W . (112) \langle p\rangle= \int dz\,dp\,pW,\qquad\sigma_p^2 = \int dz\,dp\,(p-\langle p\rangle)^2W,\qquad b_- = \int dz\,dp\,e^{-iqz/\hbar}W. \tag*{(112)} ⟨ p ⟩ = ∫ d z d p p W , σ p 2 = ∫ d z d p ( p − ⟨ p ⟩ ) 2 W , b − = ∫ d z d p e − i q z /ℏ W . (112)
Introducing the Colson current means that the field equation describes a many-electron beam response. It is not a single-electron source equation. The electron distribution supplies bunching, and the common field changes that distribution in return. With a specified FEL normalization and field phase convention one may write schematically d a / d τ = − j c b − da/d\tau=-j_cb_- d a / d τ = − j c b − , where j c j_c j c is the Colson current. Numerical coefficients and field-phase conventions must be fixed by the adopted normalization. Single-electron quantum evolution enters the collective problem through the beam distribution or density matrix and its correlations.
Increasing momentum spread can make individual recoil steps less resolvable while increasing the distribution of detunings and phase mixing. The three distinct controls are recoil resolution R q R_q R q , accumulated local phase R ϕ ∼ ∣ ϵ ∣ ℓ τ R_\phi\sim|\epsilon|\ell\tau R ϕ ∼ ∣ ϵ ∣ ℓ τ , and dephasing D \mathcal{D} D from Eq. (97) . A large spread can suppress visible quantum structure and simultaneously reduce classical FEL gain. Formal substitution ϵ → 0 \epsilon\to0 ϵ → 0 does not describe all these effects.
To see explicitly which electron information is required for bunching, consider a pure electron superposition in a periodic length L L L ,
ψ e ( z , t ) = 1 L ∑ ℓ c ℓ ( t ) e i ( p 0 − ℓ q ) z / ℏ . (113) \psi_e(z,t)=\frac{1}{\sqrt{L}}\sum_{\ell}c_\ell(t)e^{i(p_0-\ell q)z/\hbar}. \tag*{(113)} ψ e ( z , t ) = L 1 ℓ ∑ c ℓ ( t ) e i ( p 0 − ℓ q ) z /ℏ . (113)
Its Wigner function is
W = 1 L ∑ ℓ , j c ℓ c j ∗ e − i ( ℓ − j ) q z / ℏ δ [ p − p 0 + ( ℓ + j ) q 2 ] . (114) W=\frac{1}{L}\sum_{\ell,j}c_\ell c_j^*e^{-i(\ell-j)qz/\hbar}\delta\left[p-p_0+\frac{(\ell+j)q}{2}\right]. \tag*{(114)} W = L 1 ℓ , j ∑ c ℓ c j ∗ e − i ( ℓ − j ) q z /ℏ δ [ p − p 0 + 2 ( ℓ + j ) q ] . (114)
Diagonal terms give momentum populations; off-diagonal terms give spatial modulation. Integration over an integer number of periods yields
b + = ⟨ e i q z / ℏ ⟩ = ∑ j c j + 1 c j ∗ , b − = b + ∗ . (115) b_+=\langle e^{iqz/\hbar}\rangle=\sum_j c_{j+1}c_j^*,\qquad b_-=b_+^*. \tag*{(115)} b + = ⟨ e i q z /ℏ ⟩ = j ∑ c j + 1 c j ∗ , b − = b + ∗ . (115)
For a mixed electron state the products are replaced by density-matrix elements. By contrast, tracing over photons in the sharp-momentum, sharp-photon-number joint state Eq. (43) gives
ρ e = ∑ ℓ ∣ c ℓ ∣ 2 ∣ p 0 − ℓ q ⟩ ⟨ p 0 − ℓ q ∣ . (116) \rho_e=\sum_\ell|c_\ell|^2|p_0-\ell q\rangle\langle p_0-\ell q|. \tag*{(116)} ρ e = ℓ ∑ ∣ c ℓ ∣ 2 ∣ p 0 − ℓ q ⟩ ⟨ p 0 − ℓ q ∣. (116)
This particular preparation has no unconditional spatial bunching. For the initial packet, however, Eq. (84) retains coherences within each recoil branch, and the corresponding bunching is
b − ( t ) = ∑ ℓ ∫ d P χ ℓ ( P , t ) χ ℓ ∗ ( P − q , t ) . (117) b_-(t)=\sum_\ell\int dP\,\chi_\ell(P,t)\chi_\ell^*(P-q,t). \tag*{(117)} b − ( t ) = ℓ ∑ ∫ d P χ ℓ ( P , t ) χ ℓ ∗ ( P − q , t ) . (117)
For the initial Gaussian it is b − ( 0 ) = e − q 2 / ( 8 σ p 0 2 ) e − i q z c / ℏ b_-(0)=e^{-q^2/(8\sigma_{p0}^2)}e^{-iqz_c/\hbar} b − ( 0 ) = e − q 2 / ( 8 σ p 0 2 ) e − i q z c /ℏ . This may be nonzero before the interaction because the packet is spatially localized; it must be distinguished from FEL-induced modulation. The subsequent variation follows from the same Raman–Nath branch amplitudes used in Eq. (99) . An ensemble with uniformly distributed centers z c z_c z c has vanishing initial mean bunching even when each individual packet is localized. A phase-defined field, wave packets, conditional measurements or a self-consistent many-electron preparation can supply electron coherences. Joint electron–field correlations can remain important even when the reduced electron bunching vanishes. Electron wave-function modulation is therefore preparation-dependent, not an automatic interpretation of every recoil population.
Photon-exchange bound, field coupling and the pendulum
The local mismatch in Eq. (50) gives a transparent recoil bound. Recoil is negligible relative to a nonzero detuning only when
∣ ε ( 2 ℓ ± 1 ) ∣ ≪ ∣ W ∣ , ℓ ≪ ℓ r ∼ ∣ W ∣ ∣ ε ∣ . (118) |\varepsilon(2\ell\pm1)| \ll|W|,\qquad\ell\ll\ell_r \sim\frac{|W|}{|\varepsilon|}. \tag*{(118)} ∣ ε ( 2 ℓ ± 1 ) ∣ ≪ ∣ W ∣ , ℓ ≪ ℓ r ∼ ∣ ε ∣ ∣ W ∣ . (118)
The estimate suppresses an order-unity factor of two. It measures how many recoil steps can be explored before the local resonance shift is comparable to the original detuning. Close to W = 0 W = 0 W = 0 this comparison fails; the reference scale must instead be the finite-time width, gain bandwidth or coherence frequency. It must not be interpreted as a singular physical recoil constraint at exact resonance.
The dimensional neighboring photon-state matrix element is g n + 1 g\sqrt{n+1} g n + 1 : n + 1 \sqrt{n+1} n + 1 follows from a ^ † ∣ n ⟩ = n + 1 ∣ n + 1 ⟩ \hat{a}^{\dagger}|n\rangle= \sqrt{n+1}|n+1\rangle a ^ † ∣ n ⟩ = n + 1 ∣ n + 1 ⟩ , whereas absorption carries n \sqrt{n} n . It is not n + 1 n+1 n + 1 without the square root. In the dimensionless ladder its coupling is g n + 1 g\sqrt{n+1} g n + 1 . At a photon/recoil scale n r ∼ ∣ W ∣ / ∣ ε ∣ n_r \sim|W|/|\varepsilon| n r ∼ ∣ W ∣/∣ ε ∣ , introduce
Λ = g n r ∣ W ∣ ∼ g ∣ W ε ∣ . (119) \Lambda= \frac{g\sqrt{n_r}}{|W|} \sim\frac{g}{\sqrt{|W\varepsilon|}}. \tag*{(119)} Λ = ∣ W ∣ g n r ∼ ∣ W ε ∣ g . (119)
For the explicit equal-frequency mode normalization of Eq. (10) , put Δ = W / T 0 \Delta= W/T_0 Δ = W / T 0 and ω q = ε / T 0 = q 2 / ( 2 m ℏ ) \omega_q = \varepsilon/T_0 = q^2/(2m\hbar) ω q = ε / T 0 = q 2 / ( 2 m ℏ ) in the undressed nonrelativistic BR model. The pump occupation is ν ‾ u \overline{\nu}_u ν u . Then
Λ ∼ g ∣ Δ ∣ ω q = α f s λ ˉ C λ ′ ∣ ξ ∣ ν ‾ u V u ′ V r ′ ∣ Δ ∣ ω q . (120) \Lambda\sim\frac{g}{\sqrt{|\Delta|\omega_q}} = \frac{\alpha_{\mathrm{fs}}\bar{\lambda}_{C}\lambda'|\xi|\sqrt{\overline{\nu}_u}}{\sqrt{V'_uV'_r|\Delta|\omega_q}}. \tag*{(120)} Λ ∼ ∣Δ∣ ω q g = V u ′ V r ′ ∣Δ∣ ω q α fs λ ˉ C λ ′ ∣ ξ ∣ ν u . (120)
For an optical pump q = 2 ℏ ω ′ / c q = 2\hbar\omega'/c q = 2ℏ ω ′ / c , this gives
n r n m i x ∼ Λ 2 = α f s 2 λ ˉ C ( λ ′ ) 4 ∣ ξ ∣ 2 ν ‾ u 8 π 2 V u ′ V r ′ ∣ Δ ∣ . (121) \frac{n_r}{n_{\mathrm{mix}}} \sim\Lambda^2 = \frac{\alpha_{\mathrm{fs}}^2\bar{\lambda}_{C}(\lambda')^4|\xi|^2\overline{\nu}_u}{8\pi^2V'_uV'_r|\Delta|}. \tag*{(121)} n mix n r ∼ Λ 2 = 8 π 2 V u ′ V r ′ ∣Δ∣ α fs 2 λ ˉ C ( λ ′ ) 4 ∣ ξ ∣ 2 ν u . (121)
Here ω q = 8 π 2 c λ C / ( λ ′ ) 2 \omega_q = 8\pi^2 c\lambda_C/(\lambda')^2 ω q = 8 π 2 c λ C / ( λ ′ ) 2 for the optical-pump recoil q = 4 π ℏ / λ ′ q = 4\pi\hbar/\lambda' q = 4 π ℏ/ λ ′ . The ordinary Compton-wavelength form follows by replacing λ ˉ C \bar{\lambda}_C λ ˉ C with λ C / ( 2 π ) \lambda_C/(2\pi) λ C / ( 2 π ) . These expressions expose the electromagnetic coupling but also its dependence on pump occupation, overlap and radiation-mode volume. The factor ν ‾ u / V u ′ \overline{\nu}_u/V'_u ν u / V u ′ is the physical pump density, so an arbitrary normalization-volume choice does not alter the result. A laboratory relativistic or transversely dressed model must use its own kinetic curvature and coupling, rather than transplant this BR prefactor.
A numerical volume estimate requires a mode geometry. For a traveling Gaussian mode whose transverse field profile is e − r 2 / w j ′ 2 e^{-r^2/w_j'^2} e − r 2 / w j ′2 , approximately constant over a length L j ′ L'_j L j ′ , the intensity-normalized effective volume is
V j ′ = ∫ ∣ u j ( r ′ ) ∣ 2 d 2 r ′ ≃ π w j ′ 2 2 L j ′ , max ∣ u j ∣ = 1. (122) V'_j = \int|u_j(\mathbf{r}')|^2d^2r' \simeq\frac{\pi w_j'^2}{2}L'_j,\qquad\max|u_j| = 1. \tag*{(122)} V j ′ = ∫ ∣ u j ( r ′ ) ∣ 2 d 2 r ′ ≃ 2 π w j ′2 L j ′ , max ∣ u j ∣ = 1. (122)
For example, w j ′ = 10 μ m w'_j = 10\,\mathrm{\mu m} w j ′ = 10 μ m and L j ′ = 1 m m L'_j = 1\,\mathrm{mm} L j ′ = 1 mm give V j ′ ≃ 1.57 × 10 − 13 m 3 V'_j \simeq1.57 \times10^{-13}\,\mathrm{m}^3 V j ′ ≃ 1.57 × 1 0 − 13 m 3 . With equal such volumes, λ ′ = 10 n m \lambda' = 10\,\mathrm{nm} λ ′ = 10 nm and ∣ ξ ∣ = 1 |\xi| = 1 ∣ ξ ∣ = 1 , the illustrative single-quantum coupling is ∣ g u ∣ ≃ 5.38 × 10 − 2 s − 1 |g_u| \simeq5.38 \times10^{-2}\,\mathrm{s}^{-1} ∣ g u ∣ ≃ 5.38 × 1 0 − 2 s − 1 . This is a normalization example, not a reconstruction of the Kling–Giese design. A standing-wave mode adds its longitudinal intensity factor; a focused traveling mode requires the integral over its changing waist. The actual radiation-mode volume is determined by cavity or wave-packet normalization, not automatically by the electron interaction length. For a box-normalized traveling pump, changing V u ′ V'_u V u ′ and its assigned occupation together leaves ν ‾ u / V u ′ \overline{\nu}_u/V'_u ν u / V u ′ and the physical drive unchanged. Consequently no unique physical volume, or numerical Λ \Lambda Λ , follows from wavelength and interaction length alone: waist, mode profile, cavity length and pump density must also be specified.
Thus Λ ≪ 1 \Lambda\ll1 Λ ≪ 1 indicates that neighboring-state mixing remains weak near the estimated recoil boundary. This is a single-electron indicator, not a universal definition of FEL small signal.
A complementary strong-mixing scale follows from g n m i x ∼ ∣ W ∣ g\sqrt{n_{\mathrm{mix}}} \sim|W| g n mix ∼ ∣ W ∣ :
n m i x ∼ W 2 g 2 , n r n m i x ∼ Λ 2 , n ≪ min ( ∣ W ∣ ∣ ε ∣ , W 2 g 2 ) . (123) n_{\mathrm{mix}} \sim\frac{W^2}{g^2},\qquad\frac{n_r}{n_{\mathrm{mix}}} \sim\Lambda^2,\qquad n \ll\min\left(\frac{|W|}{|\varepsilon|},\frac{W^2}{g^2}\right). \tag*{(123)} n mix ∼ g 2 W 2 , n mix n r ∼ Λ 2 , n ≪ min ( ∣ ε ∣ ∣ W ∣ , g 2 W 2 ) . (123)
Since n r / n m i x ∼ Λ 2 n_r/n_{\mathrm{mix}} \sim\Lambda^2 n r / n mix ∼ Λ 2 , Λ ≪ 1 \Lambda\ll1 Λ ≪ 1 means n r ≪ n m i x n_r \ll n_{\mathrm{mix}} n r ≪ n mix : as the occupation or accessible exchange scale increases, the recoil mismatch reaches the detuning scale while the matrix element is still too small for strong mixing. Conversely, Λ ≫ 1 \Lambda\gg1 Λ ≫ 1 means n m i x ≪ n r n_{\mathrm{mix}} \ll n_r n mix ≪ n r : the coupling reaches the detuning scale before the estimated recoil boundary is reached. The statement orders two crossover scales; it does not assert that the actual dynamics necessarily populates either scale. In particular, photon occupation and net exchange order need not coincide for a strongly seeded field. These are estimates for the adopted detuned ladder. They do not replace the time-dependent perturbative condition Eq. (51) or establish classical trapping by themselves.
The classical single-electron FEL equations can be expressed, after fixing a pendulum normalization and optical phase, as
ζ ˙ = ν , ν ˙ = − ∣ a ∣ sin ( ζ + ϕ ) , H p = ν 2 2 − ∣ a ∣ cos ( ζ + ϕ ) . (124) \dot{\zeta}=\nu,\qquad\dot{\nu}=-\lvert a\rvert\sin(\zeta+\phi),\qquad H_p=\frac{\nu^2}{2}-\lvert a\rvert\cos(\zeta+\phi). \tag*{(124)} ζ ˙ = ν , ν ˙ = − ∣ a ∣ sin ( ζ + ϕ ) , H p = 2 ν 2 − ∣ a ∣ cos ( ζ + ϕ ) . (124)
Here the dot is with respect to the normalized pendulum time. The cosine Hamiltonian gives the displayed sine force with the indicated sign. A sine Hamiltonian would require the corresponding phase shift. The small-amplitude bounce frequency obeys
ω B ∼ ∣ a ∣ . (125) \omega_B\sim\sqrt{\lvert a\rvert}. \tag*{(125)} ω B ∼ ∣ a ∣ . (125)
In a coherent-field semiclassical identification the potential amplitude is proportional to g n \mathfrak{g}\sqrt{n} g n , with constants fixed by the momentum and time normalization. Hence ω B 2 \omega_B^2 ω B 2 , not ω B \omega_B ω B , scales with that potential amplitude. A trapping comparison must use the free kinetic scale appropriate to the orbit, generally involving the detuning momentum and kinetic curvature. The quantum level-mixing boundary in Eq. (123) should therefore not be equated directly with a universal classical trapping threshold.
The comparison nevertheless clarifies the two ways the elementary regime can fail: accumulated recoil changes the local resonance, or the field produces substantial nonlinear motion. Their relationship requires the full normalization connecting W W W , ε \varepsilon ε , g \mathfrak{g} g , ν \nu ν and a a a .
Canonical interaction picture and perturbative structure
The classical interaction-picture construction discussed in Ref. [33 ] reinforces the phase-space interpretation. Let H = H 0 + V H=H_0+V H = H 0 + V and define the Liouville generator by L H ρ = { H , ρ } P B \mathcal{L}_H\rho=\{H,\rho\}_{\mathrm{PB}} L H ρ = { H , ρ } PB . Choosing a manageable reference flow U 0 U_0 U 0 gives
ρ I = U 0 − 1 ρ , ∂ t ρ I = L 1 I ( t ) ρ I , L 1 I = U 0 − 1 L 1 U 0 . (126) \rho_I=U_0^{-1}\rho,\qquad\partial_t\rho_I=\mathcal{L}_{1I}(t)\rho_I,\qquad\mathcal{L}_{1I}=U_0^{-1}\mathcal{L}_1U_0. \tag*{(126)} ρ I = U 0 − 1 ρ , ∂ t ρ I = L 1 I ( t ) ρ I , L 1 I = U 0 − 1 L 1 U 0 . (126)
A time-ordered Neumann–Dyson expansion follows. This is the classical counterpart of removing the manageable detuning evolution before expanding residual recoil phases. It explains why a local mismatch organization is more useful than expanding the absolute diagonal quantity ε ℓ 2 \varepsilon\ell^2 ε ℓ 2 .
The Wigner equation similarly separates classical transport and quantum corrections,
∂ t W = ( L c l + L q ) W , (127) \partial_t\mathcal{W}=(\mathcal{L}_{\mathrm{cl}}+\mathcal{L}_q)\mathcal{W}, \tag*{(127)} ∂ t W = ( L cl + L q ) W , (127)
where L q \mathcal{L}_q L q contains the higher derivatives associated with the Moyal bracket. For a quadratic free Hamiltonian and a spatial potential, those terms are the higher odd momentum derivatives in Eq. (108) . For a nonquadratic dispersion the kinetic corrections in Eq. (107) must also be included.
A finite Dyson or Neumann polynomial need not remain an exact unitary propagator or an exact canonical-flow operator, even if the underlying dynamics has that structure. This does not mean that every finite-order density expansion necessarily violates its integral normalization: the property depends on the generator and truncation. A Magnus organization, with its exponent in the relevant Lie algebra, preserves the operator structure more directly. In the quantum problem its exponent is anti-Hermitian; in a suitable classical Hamiltonian-flow formulation its generator retains the canonical structure.
For the Raman–Nath probability expansion, order-by-order normalization requires
∑ ℓ P ℓ ( 0 ) = 1 , ∑ ℓ P ℓ ( 1 ) = 0 , ∑ ℓ P ℓ ( 2 ) = 0. (128) \sum_{\ell}P_{\ell}^{(0)}=1,\qquad\sum_{\ell}P_{\ell}^{(1)}=0,\qquad\sum_{\ell}P_{\ell}^{(2)}=0. \tag*{(128)} ℓ ∑ P ℓ ( 0 ) = 1 , ℓ ∑ P ℓ ( 1 ) = 0 , ℓ ∑ P ℓ ( 2 ) = 0. (128)
These conditions are compatible with Eq. (74) when the amplitude hierarchy is generated consistently. A finite truncation need not be positive or uniformly accurate outside the controlled domain. A structure-preserving treatment is particularly useful before iterating a perturbative channel through many cavity passages.
From single-electron dynamics to beam response
For independent incoming electrons with a normalized initial phase-space distribution f 0 ( p , θ ) f_0(p,\theta) f 0 ( p , θ ) , a linear spectral response can be averaged as
Γ ‾ ± ( Δ ) = N e ∫ d p d θ f 0 ( p , θ ) Γ ± ( p , θ ; Δ ) . (129) \overline{\Gamma}_{\pm}(\Delta)=N_e\int dp\,d\theta\,f_0(p,\theta)\Gamma_{\pm}(p,\theta;\Delta). \tag*{(129)} Γ ± ( Δ ) = N e ∫ d p d θ f 0 ( p , θ ) Γ ± ( p , θ ; Δ ) . (129)
This is useful for initial response and classical correspondence. It is not a claim that electrons remain unmodified in low gain. The outgoing distribution acquires perturbative changes at the order at which energy modulation and bunching are calculated.
Even a linearized high-gain calculation must retain the self-consistent field–distribution coupling. A fixed-distribution convolution supplies an elementary beam-averaged kernel; sustained collective amplification requires a kinetic closure or a many-electron quantum hierarchy. The relevant dynamical information includes the field amplitude and photon number, electron bunching, mean momentum, momentum variance, and correlations that are not determined by those moments alone.
A quantum treatment averages incoming electron density operators and transition channels. Electron shot noise, vacuum emission and cavity reservoir noise must be included consistently, once each. An extra phenomenological spontaneous seed can double count feeding already present in a quantum interaction map. The independent classical shot-noise baseline is discussed in Appendix § 1.15 ; it remains essential when deciding whether a measured fluctuation is a quantum signature.
Quantum cavity evolution of the FEL oscillator
A resonator stores the field while fresh electron bunches repeatedly interact with it. Its stationary state depends on pumping, loss, saturation and mode overlap; it is not obtained by multiplying a single-electron recoil correction by the number of passages.
The cavity research initiated by Walther provides a particularly relevant experimental precedent. Meschede, Walther and Müller demonstrated photon exchange between individual Rydberg atoms and a superconducting microwave cavity in the one-atom maser [45 ] . Rempe, Walther and Klein subsequently observed quantum collapse and revival, with velocity selection controlling the interaction time [46 ] . These experiments and the cavity-QED review by Walther and collaborators [47 ] establish the value of preparing the incident particle, controlling its transit, storing a small optical or microwave occupation, and inferring field statistics from particle dynamics. The Jaynes–Cummings description [48 ] supplies the elementary atom–field model: occupation-dependent exchange frequencies can dephase and later rephase. Its formal relation to an isolated electron recoil pair is useful, but the atomic internal transition and the free-electron momentum ladder have different preparation and coupling requirements. Collapse and revival are therefore a possible target only after coherence, interaction-time uniformity and cavity lifetime have been established; they are not implied by a stationary FEL photon distribution.
Let ρ m \rho_m ρ m be the intracavity field density operator before bunch m m m , and ρ e , i n ( m ) \rho_{e,\mathrm{in}}^{(m)} ρ e , in ( m ) the state of the fresh incoming electrons. The single-pass field channel is
E m ( ρ m ) = Tr e [ U m ( ρ m ⊗ ρ e , i n ( m ) ) U m † ] . (130) \mathcal{E}_m(\rho_m)=\operatorname{Tr}_e\left[U_m\left(\rho_m\otimes\rho_{e,\mathrm{in}}^{(m)}\right)U_m^\dagger\right]. \tag*{(130)} E m ( ρ m ) = Tr e [ U m ( ρ m ⊗ ρ e , in ( m ) ) U m † ] . (130)
The outgoing electrons are traced out because they leave the cavity. Their recoil and correlations with the light affect the field channel, but their outgoing state is not reused as the next incoming bunch. For independently prepared incoming bunches the exact channel is completely positive and trace preserving.
Including the cavity phase, losses and pulse timing gives
ρ m + 1 = L η , ϕ c , δ [ E m ( ρ m ) ] . (131) \rho_{m+1}=\mathcal{L}_{\eta,\phi_c,\delta}\left[\mathcal{E}_m(\rho_m)\right]. \tag*{(131)} ρ m + 1 = L η , ϕ c , δ [ E m ( ρ m ) ] . (131)
The loss channel includes the appropriate environmental modes. For a single port with fractional power loss 0 ≤ η ≤ 1 0 \le\eta\le1 0 ≤ η ≤ 1 per passage through that loss element, its operator representation is
a ^ a f t e r = 1 − η a ^ b e f o r e + η v ^ , (132) \hat{a}_{\mathrm{after}}=\sqrt{1-\eta}\,\hat{a}_{\mathrm{before}}+\sqrt{\eta}\,\hat{v}, \tag*{(132)} a ^ after = 1 − η a ^ before + η v ^ , (132)
where v ^ \hat{v} v ^ is an independent reservoir mode. Vacuum noise is consequently an operator contribution. An equivalent stochastic amplitude formulation must state its phase-space representation and noise prescription.
A selected single-mode model is an appropriate reduction when timing, spatial overlap and spectral selection support it. Finite bunch duration, slippage, radiation lethargy and longitudinal mode competition can instead require an envelope description. The long-bunch, low-gain pulse-propagation review in Ref. [49 ] provides the relevant envelope treatment. The evolution-operator and time-ordering methods of Ref. [36 ] provide its mathematical context, while Ref. [50 ] discusses high-gain oscillator pulse propagation and saturation. These extensions do not invalidate the selected-mode reduction when its assumptions hold. In a classical envelope approximation,
A m + 1 ( ζ ) = 1 − η e i ϕ c T δ L { U F E L [ A m , f e , i n ( m ) ] } ( ζ ) , T δ L = e − δ ζ ∂ ζ . (133) A_{m+1}(\zeta)=\sqrt{1-\eta}\,e^{i\phi_c}T_{\delta L}\{\mathcal{U}_{\mathrm{FEL}}[A_m,f_{e,\mathrm{in}}^{(m)}]\}(\zeta),\qquad T_{\delta L}=e^{-\delta_\zeta\partial_\zeta}. \tag*{(133)} A m + 1 ( ζ ) = 1 − η e i ϕ c T δ L { U FEL [ A m , f e , in ( m ) ]} ( ζ ) , T δ L = e − δ ζ ∂ ζ . (133)
Here η \eta η is the total fractional intracavity power lost in one round trip, including output coupling and internal losses, with 0 ≤ η < 1 0 \le\eta< 1 0 ≤ η < 1 for an operating cavity. It is dimensionless and differs from an amplitude-loss fraction: the amplitude transmission is 1 − η \sqrt{1-\eta} 1 − η . The symbol δ ζ \delta_\zeta δ ζ is the dimensionless relative pulse displacement produced by the cavity mismatch δ L \delta L δ L , with its conversion determined by the chosen longitudinal coordinate. Discrete-electron noise can enter a stochastic classical treatment; a quantum treatment derives its fluctuations from the channel rather than appending an unspecified random field.
An equilibrium pulse satisfies a functional fixed-point condition, up to an overall phase. If G ( I ) G(I) G ( I ) is the single-pass fractional intensity gain, its scalar intensity projection is
( 1 − η ) [ 1 + G ( I e q ) ] = 1 , G ( I e q ) = η 1 − η ≃ η . (134) (1-\eta)[1+G(I_{\mathrm{eq}})]=1,\qquad G(I_{\mathrm{eq}})=\frac{\eta}{1-\eta}\simeq\eta. \tag*{(134)} ( 1 − η ) [ 1 + G ( I eq )] = 1 , G ( I eq ) = 1 − η η ≃ η . (134)
The final approximation applies to small loss. This determines only a mean-intensity condition; photon statistics require the stationary density operator.
Comparison with Kling and Giese
An innovative route toward dedicated quantum-FEL experiments
Kling and Giese [24 ] propose an interesting and innovative quantum FEL oscillator: a counter-propagating optical undulator produces well-separated recoil levels, while successive fresh electron bunches amplify radiation stored in a resonator. The short interaction region reduces the accumulated effects of space charge and unwanted spontaneous emission compared with a long single-pass device. Their proposal gives a constructive perspective for dedicated experiments on recoil-selective electron–photon exchange and radiation statistics. The following discussion identifies the conditions needed to develop that perspective; it does not imply that the proposed quantum regime is impossible.
The resonant pair in the BR frame is ∣ p , n ⟩ ↔ ∣ p − q , n + 1 ⟩ |p,n\rangle\leftrightarrow|p-q,n+1\rangle ∣ p , n ⟩ ↔ ∣ p − q , n + 1 ⟩ , with p ≃ q / 2 p \simeq q/2 p ≃ q /2 . For recoil momentum q = 2 ℏ k q=2\hbar k q = 2ℏ k , the recoil frequency and two-state selectivity parameter are
ω r = q 2 2 m ℏ , α n = g n + 1 ω r ≪ 1. (135) \omega_r=\frac{q^2}{2m\hbar},\qquad\alpha_n=\frac{g\sqrt{n+1}}{\omega_r}\ll1. \tag*{(135)} ω r = 2 m ℏ q 2 , α n = ω r g n + 1 ≪ 1. (135)
Here g g g and ω r \omega_r ω r are BR-frame angular frequencies and T T T below is the interaction duration in that frame. The optical-undulator pump is represented by a prescribed classical amplitude, consistent with the undepleted, highly occupied coherent-pump limit discussed in Section VI. It must be distinguished from the stored radiation, whose stationary state need not be coherent. A photon-number-state pump represents a different preparation; artificial recoil-free antibunching from that choice is excluded by coherent-pump averaging [39 ] .
What the stationary photon distribution includes
The two-state transition probability is
Ω n ( p ) = g 2 ( n + 1 ) + ω r 2 ( p / q − 1 / 2 ) 2 , P e m ( n , p ) = g 2 ( n + 1 ) T 2 sinc 2 [ Ω n ( p ) T ] . (136) \begin{aligned}
\Omega_n(p) &= \sqrt{g^2(n+1)+\omega_r^2(p/q-1/2)^2},\\
P_{\mathrm{em}}(n,p) &= g^2(n+1)T^2\operatorname{sinc}^2[\Omega_n(p)T]. \tag*{(136)}
\end{aligned} Ω n ( p ) P em ( n , p ) = g 2 ( n + 1 ) + ω r 2 ( p / q − 1/2 ) 2 , = g 2 ( n + 1 ) T 2 sinc 2 [ Ω n ( p ) T ] . (136)
The n + 1 n+1 n + 1 factor includes the vacuum contribution and stimulated enhancement by the stored photons. The sinc factor expresses finite-time resonance acceptance and coherent exchange; it does not mean that only spontaneous radiation is stored. Reabsorption within the resonant pair is part of the same unitary evolution.
Under their independent-event pumping and cavity-damping assumptions, the stationary probabilities are
P n = P 0 ∏ j = 1 n [ ϑ 2 ∫ d p ρ e ( p ) sinc 2 [ Ω j − 1 ( p ) T ] ] , (137) P_n=P_0\prod_{j=1}^{n}\left[\vartheta^2\int dp\,\rho_e(p)\operatorname{sinc}^2[\Omega_{j-1}(p)T]\right], \tag*{(137)} P n = P 0 j = 1 ∏ n [ ϑ 2 ∫ d p ρ e ( p ) sinc 2 [ Ω j − 1 ( p ) T ] ] , (137)
ϑ = g T N a , N a = N e Q ω L τ i n j . \vartheta=gT\sqrt{N_a},\qquad N_a=\frac{N_eQ}{\omega L\tau_{\mathrm{inj}}}. ϑ = g T N a , N a = ω L τ inj N e Q .
N e N_e N e denotes electrons per bunch, Q Q Q the cavity quality factor and ρ e ( p ) \rho_e(p) ρ e ( p ) the normalized entrance momentum density. Thus entrance energy averaging is already present. The result is a stationary distribution after repeated pumping and loss, not the distribution immediately after one round trip. In the birth–death balance, the factor j j j in emission from j − 1 j-1 j − 1 photons cancels the factor j j j in cavity loss. Its absence from the displayed product does not remove stimulated emission. Their supplementary derivation [51 ] traces out each outgoing electron; additional initial electron coherences or collective bunch correlations require the appropriate joint channel.
The diagnostic is F = Var ( n ) / ⟨ n ⟩ F=\operatorname{Var}(n)/\langle n\rangle F = Var ( n ) / ⟨ n ⟩ . Near threshold, their resonant small-signal estimate is F q ≃ 1 / δ F_q\simeq1/\delta F q ≃ 1/ δ , where δ = ( G − ω L τ i n j / Q ) / G \delta=(G-\omega L\tau_{\mathrm{inj}}/Q)/G δ = ( G − ω L τ inj / Q ) / G and G = N e ( g T ) 2 G=N_e(gT)^2 G = N e ( g T ) 2 . It is therefore super-Poissonian for 0 < δ ≪ 1 0<\delta\ll1 0 < δ ≪ 1 , although narrower than their classical small-recoil comparison. Sub-Poissonian regions, F < 1 F<1 F < 1 , occur in suitable nonlinear operating regimes of their model. Reduced excess fluctuations and nonclassical counting statistics should consequently be distinguished.
Reported beam parameters and the available acceptance
Tables 1–2 of the proposal provide the quantitative starting point [24 ] :
γ 0 51.8 L 1.14 m m λ W 1.064 μ m λ L 1 A ˚ n e 6.39 × 10 22 m − 3 I p 8.9 A σ e 0.68 μ m τ e 1.2 p s Q b 11 p C ϵ n 0.0082 m m m r a d σ ϵ , 0 7.5 × 10 − 5 G 0.1 \begin{aligned}
\gamma_0&\quad51.8 & L&\quad1.14\ \mathrm{mm}\\
\lambda_W&\quad1.064\ \mu\mathrm{m} & \lambda_L&\quad1\ \text{\AA}\\
n_e&\quad6.39\times10^{22}\ \mathrm{m}^{-3} & I_p&\quad8.9\ \mathrm{A}\\
\sigma_e&\quad0.68\ \mu\mathrm{m} & \tau_e&\quad1.2\ \mathrm{ps}\\
Q_b&\quad11\ \mathrm{pC} & \epsilon_n&\quad0.0082\ \mathrm{mm\,mrad}\\
\sigma_{\epsilon,0}&\quad7.5\times10^{-5} & G&\quad0.1
\end{aligned} γ 0 λ W n e σ e Q b σ ϵ , 0 51.8 1.064 μ m 6.39 × 1 0 22 m − 3 0.68 μ m 11 pC 7.5 × 1 0 − 5 L λ L I p τ e ϵ n G 1.14 mm 1 A ˚ 8.9 A 1.2 ps 0.0082 mm mrad 0.1
The quoted spread is 0.075 per mille, not 0.075 percent. Numerical width estimates below interpret it as rms and interpret σ e \sigma_e σ e as a one-plane rms waist size. The reported bunch-duration scale gives ℓ b ≃ c τ e = 0.360 m m \ell_b\simeq c\tau_e=0.360\ \mathrm{mm} ℓ b ≃ c τ e = 0.360 mm , about one-third of L L L ; its precise profile convention must be retained in a detailed overlap calculation. The optical pulse duration 7.6 ps corresponds to 2.28 mm, approximately 2 L 2L 2 L for a counterpropagating pulse, and the Rayleigh length is 4.4 mm. The article’s slippage estimate ( L / λ W ) λ L ≃ 0.107 μ m (L/\lambda_W)\lambda_L\simeq0.107\ \mu\mathrm{m} ( L / λ W ) λ L ≃ 0.107 μ m is much smaller than the bunch length.
The current and density are consistent with the transverse Gaussian convention
n e , 0 = I p 2 π e β c σ x σ y , I p = Q b 2 π σ t for a Gaussian time profile . (138) n_{e,0}=\frac{I_p}{2\pi e\beta c\sigma_x\sigma_y},\qquad
I_p=\frac{Q_b}{\sqrt{2\pi}\sigma_t}\quad\text{for a Gaussian time profile}. \tag*{(138)} n e , 0 = 2 π e β c σ x σ y I p , I p = 2 π σ t Q b for a Gaussian time profile . (138)
The second relation is a profile-specific identity, not an identification of the quoted τ e \tau_e τ e with σ t \sigma_t σ t . The reported density, radius and current give I p ≃ 8.9 A I_p\simeq8.9\ \mathrm{A} I p ≃ 8.9 A without assuming that identification.
Writing N W = L / λ W ≃ 1071 N_W=L/\lambda_W\simeq1071 N W = L / λ W ≃ 1071 for the laboratory optical period count, their finite-time electron-energy acceptance is is
Γ = λ W 4 π L ≃ 7.45 × 10 − 5 , μ ϵ , W = 4 N W σ ϵ , 0 ≃ 0.321 , σ ϵ , 0 Γ ≃ 1. (139) \Gamma= \frac{\lambda_W}{4\pi L} \simeq7.45 \times10^{-5}, \qquad\mu_{\epsilon,W} = 4N_W\sigma_{\epsilon,0} \simeq0.321, \qquad\frac{\sigma_{\epsilon,0}}{\Gamma} \simeq1. \tag*{(139)} Γ = 4 π L λ W ≃ 7.45 × 1 0 − 5 , μ ϵ , W = 4 N W σ ϵ , 0 ≃ 0.321 , Γ σ ϵ , 0 ≃ 1. (139)
For comparison with the conventional magnetic-undulator detuning kernel, the counterpropagating wave has effective period count N e f f ≃ 2 N W N_{\mathrm{eff}} \simeq2N_W N eff ≃ 2 N W ; hence μ ϵ , e f f ≃ 0.643 \mu_{\epsilon,\mathrm{eff}} \simeq0.643 μ ϵ , eff ≃ 0.643 and the accumulated rms detuning phase is 2 σ ϵ , 0 / Γ ≃ 2.0 2\sigma_{\epsilon,0}/\Gamma\simeq2.0 2 σ ϵ , 0 /Γ ≃ 2.0 . These conventions must not be interchanged. At 1 Å the photon energy is 12.40 keV and
r = ℏ ω L γ 0 m c 2 ≃ 4.68 × 10 − 4 , σ ϵ , 0 r ≃ 0.16. (140) r = \frac{\hbar\omega_L}{\gamma_0mc^2} \simeq4.68 \times10^{-4}, \qquad\frac{\sigma_{\epsilon,0}}{r} \simeq0.16. \tag*{(140)} r = γ 0 m c 2 ℏ ω L ≃ 4.68 × 1 0 − 4 , r σ ϵ , 0 ≃ 0.16. (140)
Entrance spread alone therefore permits recoil separation, while already occupying a substantial part of the narrower resonance acceptance. A Gaussian, weak-field, on-resonance average of sinc 2 [ ( δ γ / γ 0 ) / Γ ] \operatorname{sinc}^2[(\delta\gamma/\gamma_0)/\Gamma] sinc 2 [( δ γ / γ 0 ) /Γ ] gives about 0.76 of the monoenergetic response. This is a diagnostic estimate, not a recalculated cavity threshold.
The reported density also reproduces the proposed Coulomb phase:
Ω p 2 = e 2 n e ϵ 0 m γ 0 3 , μ Q = Ω p L c ≃ 0.145. (141) \Omega_p^2 = \frac{e^2 n_e}{\epsilon_0 m \gamma_0^3}, \qquad\mu_Q = \frac{\Omega_p L}{c} \simeq0.145. \tag*{(141)} Ω p 2 = ϵ 0 m γ 0 3 e 2 n e , μ Q = c Ω p L ≃ 0.145. (141)
This is a collective-response parameter, not an rms energy width to add in quadrature. Its gain-dilution role can resemble inhomogeneous broadening [37 ] ; the actual induced chirp requires the bunch field and geometry. The broader high-gain comparison is given in Appendix § 1.14 .
Divergence, overlap and intensity-dependent redistribution
External focusing can place a waist inside an otherwise unfocused interaction region. The quoted emittance and radius give, neglecting collective forces,
σ x ′ = ϵ n β γ 0 σ e ≃ 0.233 m r a d , β ∗ ≃ γ 0 σ e 2 ϵ n = 2.92 m m , σ x ( z ) = σ e 1 + z 2 / β ∗ 2 . (142) \sigma_{x'} = \frac{\epsilon_n}{\beta\gamma_0\sigma_e} \simeq0.233\ \mathrm{mrad}, \qquad\beta_* \simeq\frac{\gamma_0\sigma_e^2}{\epsilon_n} = 2.92\ \mathrm{mm}, \qquad\sigma_x(z) = \sigma_e\sqrt{1+z^2/\beta_*^2}. \tag*{(142)} σ x ′ = β γ 0 σ e ϵ n ≃ 0.233 mrad , β ∗ ≃ ϵ n γ 0 σ e 2 = 2.92 mm , σ x ( z ) = σ e 1 + z 2 / β ∗ 2 . (142)
For a centred waist, the size grows by only 1.9% at either end of the interaction. Free drift preserves the angular distribution; it develops position–angle correlations rather than generating additional angular spread. A density-weighted hourglass estimate gives H e = ( 2 β ∗ / L ) arctan [ L / ( 2 β ∗ ) ] ≃ 0.988 H_e = (2\beta_*/L)\arctan[L/(2\beta_*)] \simeq0.988 H e = ( 2 β ∗ / L ) arctan [ L / ( 2 β ∗ )] ≃ 0.988 . This is not a universal gain correction: the actual electron–optical-mode overlap, including the stored X-ray mode, must be evaluated.
Angular detuning is more significant. In the article’s optical-undulator convention, for a fixed axial radiation mode its equivalent electron-energy detuning is approximately
d θ = γ 0 2 ( x ′ 2 + y ′ 2 ) 2 ( 1 + a 0 2 ) , a 0 = 0.0944. (143) d_\theta= \frac{\gamma_0^2(x'^2+y'^2)}{2(1+a_0^2)}, \qquad a_0 = 0.0944. \tag*{(143)} d θ = 2 ( 1 + a 0 2 ) γ 0 2 ( x ′2 + y ′2 ) , a 0 = 0.0944. (143)
For independent equal Gaussian angles in the two planes, its mean and standard deviation are both about 1.44 × 10 − 4 1.44 \times10^{-4} 1.44 × 1 0 − 4 . Retuning can remove the mean, but not the angular distribution. Combining its variance with independent entrance energy variance gives a diagnostic width 1.62 × 10 − 4 ≃ 2.2 Γ 1.62 \times10^{-4} \simeq2.2\Gamma 1.62 × 1 0 − 4 ≃ 2.2Γ . The angular distribution is asymmetric, so the full joint average, rather than a Gaussian width alone, is required. These estimates identify a demanding acceptance margin; they do not prove that recoil is unresolved or that the proposed state cannot be reached.
The growing intracavity field also changes the outgoing electron distribution. Ordinary low-gain oscillator theory gives an induced relative energy spread proportional to the square root of local stored intensity [52 ] ; its small-signal expression must not be extrapolated uncritically to the recoil-resolved regime. In the KG model, stimulated recoil back-action is already included in Eq. (136) . For an initially sharp resonant electron, the two outgoing energy populations yield
σ ϵ , i n d = r P e m ( 1 − P e m ) . (144) \sigma_{\epsilon,\mathrm{ind}} = r\sqrt{P_{\mathrm{em}}(1-P_{\mathrm{em}})}. \tag*{(144)} σ ϵ , ind = r P em ( 1 − P em ) . (144)
Fresh bunches do not carry this outgoing spread from one round trip to the next, but each bunch experiences the field accumulated in the cavity. The two-state approximation must remain valid throughout that buildup. For their near-threshold example, δ = 0.01 \delta= 0.01 δ = 0.01 gives
n s s ≃ 3 δ N e G = 2.07 × 10 7 , N r t ≃ ln ( 3 N e δ / G 2 ) G δ ≃ 1.9 × 10 4 . (145) n_{\mathrm{ss}} \simeq\frac{3\delta N_e}{G} = 2.07 \times10^{7}, \qquad N_{\mathrm{rt}} \simeq\frac{\ln(3N_e\delta/G^2)}{G\delta} \simeq1.9 \times10^{4}. \tag*{(145)} n ss ≃ G 3 δ N e = 2.07 × 1 0 7 , N rt ≃ G δ ln ( 3 N e δ / G 2 ) ≃ 1.9 × 1 0 4 . (145)
At 10 MHz this is approximately 1.9 ms. The resonant estimate g T n s s ≃ 3 δ = 0.173 gT\sqrt{n_{\mathrm{ss}}} \simeq\sqrt{3}\delta= 0.173 g T n ss ≃ 3 δ = 0.173 gives P e m ≃ 0.030 P_{\mathrm{em}} \simeq0.030 P em ≃ 0.030 and σ ε , i n d ≃ 8.0 × 10 − 5 \sigma_{\varepsilon,\mathrm{ind}} \simeq8.0 \times10^{-5} σ ε , ind ≃ 8.0 × 1 0 − 5 , comparable with the entrance spread and acceptance, but below the recoil. Near-threshold equilibrium therefore need not imply deep saturation or recoil washout. It remains sensitive to beam averaging and optical overlap; the quoted buildup time and equilibrium should be reassessed with those effects.
For independent contributions variances add, while correlated contributions also require covariance terms. Adding the illustrative entrance, angular and outgoing induced widths in quadrature gives about 1.8 × 10 − 4 1.8 \times10^{-4} 1.8 × 1 0 − 4 , or 2.4 Γ 2.4\Gamma 2.4Γ . This is only a comparison of scales: the induced distribution is conditional on the input and the field, and inserting that outgoing width back into the entrance integral would double-count stimulated dynamics. A joint electron–field calculation should propagate the redistribution once. The present scale estimates therefore motivate refinement rather than a categorical negative conclusion.
Experimental signature and constructive perspective
The decisive test is photon-number statistics in a defined detected mode or pulse. Calibrated photon counting, or an equivalent calibrated pulse-energy measurement, can determine F F F . For independent loss and detection with net efficiency η \eta η ,
F d e t = 1 + η ( F c a v − 1 ) . (146) F_{\mathrm{det}} = 1 + \eta(F_{\mathrm{cav}} - 1). \tag*{(146)} F det = 1 + η ( F cav − 1 ) . (146)
Loss moves F F F toward unity; technical fluctuations, detector response and multimode collection must be characterized separately. Sub-Poissonian counting would establish nonclassical radiation, whereas reduced super-Poissonian fluctuations alone would not. Attribution to recoil-selective dynamics requires scans of detuning, energy spread, divergence and stored intensity against compatible quantum and classical predictions. A very weak interaction minimizes induced redistribution but may also weaken the statistical signal; the useful experimental window must satisfy both gain–loss balance and observability.
The proposal’s importance lies in opening a concrete route to such dedicated tests, rather than in already proving experimental observability. Recoil separation is plausible for the reported entrance spread, but the combined beam-quality and intensity-dependent effects leave a demanding resonance margin. These criticalities should encourage further calculations and controlled experiments. The authors’ innovative oscillator approach is a positive contribution toward separating genuine quantum signatures from classical beam and cavity fluctuations.
Kapitza–Dirac diffraction and a complementary coincidence experiment
The Kapitza–Dirac effect offers a direct physical bridge between matter-wave diffraction and the microscopic exchange underlying an optical FEL. Kapitza and Dirac proposed electron scattering by standing light waves [53 ] ; resolved electron diffraction was later demonstrated by Freimund, Afatoni and Batelaan [54 ] , with the development and interpretation reviewed by Batelaan [55 ] . Its importance here is that a periodic optical interaction produces discrete electron momentum orders. This establishes quantum electron diffraction, while a sufficiently strong prescribed classical optical potential can already describe that diffraction. It does not alone establish nonclassical radiation or quantum FEL amplification.
A complementary development of the Kling–Giese experimental perspective is a cavity supporting two quantized counterpropagating optical modes, crossed transversely by individually detected electrons. Let the light propagate along ± z \pm z ± z , the electron transport predominantly along x x x , and p 0 p_0 p 0 denote the entrance z z z -momentum, scanned by a small incidence-angle change. In an effective nonrelativistic ponderomotive reduction,
H K D , i n t = ℏ g ( a − † a + e 2 i k z + a + † a − e − 2 i k z ) , q = 2 ℏ k . (147) H_{\mathrm{KD,int}} = \hbar g\left(a_{-}^{\dagger}a_{+}e^{2ikz} + a_{+}^{\dagger}a_{-}e^{-2ikz}\right), \qquad q = 2\hbar k. \tag*{(147)} H KD , int = ℏ g ( a − † a + e 2 ik z + a + † a − e − 2 ik z ) , q = 2ℏ k . (147)
The two exchange directions correlate p 0 → p 0 ± q p_0 \to p_0 \pm q p 0 → p 0 ± q with a photon transferred into the opposite travelling optical mode. The small-angle electron deflection is approximately Δ θ = ± q / p x \Delta\theta= \pm q/p_x Δ θ = ± q / p x . A ring cavity with independently accessible travelling modes is a candidate geometry; output-mode mixing, cavity backscattering and delayed leakage must be calibrated before photon counts can be assigned to directional transfer. The relativistic implementation requires the appropriate reduction of the laboratory Hamiltonian rather than direct use of the nonrelativistic coupling.
The proposed diagnostic is electron deflection in coincidence with the optical outcome. If P + ( p 0 ) P_+(p_0) P + ( p 0 ) and P − ( p 0 ) P_-(p_0) P − ( p 0 ) are the weak single-exchange probabilities for increasing and decreasing the − - − mode occupation, respectively, the mean transfer is
A − ( p 0 ) ≡ ⟨ n − ⟩ o u t − ⟨ n − ⟩ i n = P + ( p 0 ) − P − ( p 0 ) , G − ( p 0 ) = A − ( p 0 ) ⟨ n − ⟩ i n . (148) \mathcal{A}_-(p_0) \equiv\langle n_-\rangle_{\mathrm{out}}-\langle n_-\rangle_{\mathrm{in}} = P_+(p_0)-P_-(p_0), \qquad\mathcal{G}_-(p_0) = \frac{\mathcal{A}_-(p_0)}{\langle n_-\rangle_{\mathrm{in}}}. \tag*{(148)} A − ( p 0 ) ≡ ⟨ n − ⟩ out − ⟨ n − ⟩ in = P + ( p 0 ) − P − ( p 0 ) , G − ( p 0 ) = ⟨ n − ⟩ in A − ( p 0 ) . (148)
The first equality to the probability difference is a single-exchange approximation. A − \mathcal{A}_- A − is the mode-transfer gain curve; fractional gain G − \mathcal{G}_- G − requires a nonzero seed, whereas vacuum input gives a photon yield. For the ideal interaction, conservation of p z − q n − p_z-qn_- p z − q n − yields the exact balance Δ ⟨ p z ⟩ = q Δ ⟨ n − ⟩ \Delta\langle p_z\rangle=q\Delta\langle n_-\rangle Δ ⟨ p z ⟩ = q Δ ⟨ n − ⟩ , before leakage or other momentum-changing processes. With equal frequencies and symmetrically populated input modes, the two directional probabilities are interchanged by p 0 → − p 0 p_0 \to-p_0 p 0 → − p 0 . The mean transfer is then antisymmetric and vanishes at p 0 = 0 p_0=0 p 0 = 0 , even when both scattering channels are populated. For signed exchange order ℓ = + 1 , − 1 , 0 \ell=+1,-1,0 ℓ = + 1 , − 1 , 0 ,
⟨ ℓ ⟩ = P + − P − , ⟨ ℓ 2 ⟩ = P + + P − . (149) \langle\ell\rangle=P_+-P_-, \qquad\langle\ell^2\rangle=P_++P_-. \tag*{(149)} ⟨ ℓ ⟩ = P + − P − , ⟨ ℓ 2 ⟩ = P + + P − . (149)
Coincidences and fluctuations can thus reveal microscopic exchange hidden by a zero mean. The averaging issue identified by Carmesin et al. [43 ] supplies the motivation for this diagnostic: agreement of first moments is a weaker test than agreement of joint outcomes. The measured transfer still requires convolution with the entrance momentum distribution; conditional detection does not remove unresolved beam uncertainty.
For equal optical frequencies the ideal exchange conserves total optical energy: one mode grows at the expense of the other. Electron-powered optical amplification requires a suitable frequency difference and Doppler resonance. Low mode occupation limits perturbation, but must coexist with sufficient single-photon coupling and detection efficiency. An incoming number state is already a nonclassical resource; a weak coherent-state control is needed to distinguish transferred input nonclassicality from effects generated in the interaction. Directional population correlations identify exchange but do not by themselves certify entanglement, which requires a coherence-sensitive witness. The proposal is therefore a constructive route to controlled electron–photon quantum dynamics, informed by Carmesin’s averaging analysis and Walther’s preparation-and-readout strategy, with experimental feasibility still to be established.
Stimulated Čerenkov exchange as an experimental precedent
The observation of stimulated quantum Čerenkov exchange by Nehemia and collaborators provides an encouraging experimental precedent [56 ] . Electrons grazing a dielectric prism interact with the evanescent optical field produced by total internal reflection. Matching the longitudinal optical phase velocity to the electron velocity permits coherent interaction over hundreds of micrometres, producing resolved electron energy sidebands separated by ℏ ω \hbar\omega ℏ ω . At a representative kinetic energy of 207.2 keV and optical wavelength of 730 nm, the photon spacing is approximately 1.7 eV. The reported spectrometer resolution is approximately 0.1 eV; importantly, the effective resolution of individual sidebands is limited by the electron zero-loss peak width, approximately 1.1 eV in most measurements. These results demonstrate that adequate preparation, synchronism and energy resolution can make stimulated quantum electron dynamics experimentally accessible.
This perspective complements the Kapitza–Dirac and cavity proposals discussed above. The Čerenkov geometry supplies longitudinal synchronism through the dielectric optical field, whereas the proposed two-mode Kapitza–Dirac experiment measures transverse momentum exchange. Resolved electron sidebands support the quantum description of the electron, but can also be obtained with a prescribed coherent optical field; their observation does not by itself establish nonclassical photon statistics or recoil-resolved FEL oscillation. Extending such measurements to electron–photon coincidences would therefore address a further question. The reported sensitivity of the interaction strength to electron energy, and the trade-off between interaction length and angular spreading near the surface, reinforce the importance of characterized beam preparation. The achievement nevertheless supports the constructive conclusion that dedicated geometries and suitable sensitivity can expose microscopic quantum exchange.
The connection with Čerenkov FELs also has an established theoretical background. Walsh, Johnson, Dattoli and Renieri compared undulator and Čerenkov free-electron lasers in the Compton regime for short-pulse relativistic electrons. tron beams [57 ] . This comparison places the present experimental perspective within the broader study of stimulated electron–radiation exchange, while retaining the distinct synchronism mechanisms of the two devices.
Conclusions
The analysis shows that recoil observability depends on accumulated transition phases, electron and optical preparation, and the acceptance of the measured interaction. Neither spontaneous startup nor excess photon variance alone establishes a nonclassical FEL signature.
Cumulative recoil and the Bloch–Nordsieck interpretation
The connection with Bloch and Nordsieck [32 ] should be understood at the level of the prescribed-current, no-recoil approximation. An individual photon transfer may be small while its accumulation over many transfers changes the electron motion appreciably. Accordingly, the controlling quantity involves both the elementary recoil and the populated exchange orders. This is a physical restriction on the approximation, rather than a consequence of the infrared cancellation theorem itself.
In the notation of this article, ε \varepsilon ε is the dimensionless Raman–Nath recoil coefficient of Eq. (46) , not the fractional laboratory photon energy. The neighboring-state mismatch of Eq. (50) is
δ ℓ ± = ε ( 2 ℓ ± 1 ) , ∣ δ ℓ ± τ ∣ ≪ 1. (150) \delta_{\ell}^{\pm} = \varepsilon(2\ell\pm1), \qquad\lvert\delta_{\ell}^{\pm}\tau\rvert\ll1. \tag*{(150)} δ ℓ ± = ε ( 2 ℓ ± 1 ) , ∣ δ ℓ ± τ ∣ ≪ 1. (150)
The small-phase condition must hold over the significantly occupied ladder throughout the interaction. For large occupied ∣ ℓ ∣ \lvert\ell\rvert ∣ ℓ ∣ , its characteristic dependence is ∣ ε ℓ τ ∣ ≪ 1 \lvert\varepsilon\ell\tau\rvert\ll1 ∣ ε ℓ τ ∣ ≪ 1 , up to a factor of two. At a fixed normalized exit time of order unity this is the origin of the useful shorthand ∣ ε ℓ ∣ ≪ 1 \lvert\varepsilon\ell\rvert\ll1 ∣ ε ℓ ∣ ≪ 1 . Comparing a recoil shift with a resonance acceptance also requires both quantities in the same frequency or detuning normalization; ε ℓ \varepsilon\ell ε ℓ cannot be compared directly with the Pierce parameter without that conversion.
For a static magnetic undulator, the fractional laboratory energy transfer per resonant radiation photon is
η r e c = ℏ ω L E e , E e = γ m c 2 , Δ E e E e ≃ − ℓ η r e c . (151) \eta_{\mathrm{rec}} = \frac{\hbar\omega_L}{E_e}, \qquad E_e = \gamma m c^2, \qquad\frac{\Delta E_e}{E_e} \simeq-\ell\eta_{\mathrm{rec}}. \tag*{(151)} η rec = E e ℏ ω L , E e = γ m c 2 , E e Δ E e ≃ − ℓ η rec . (151)
Here ℓ \ell ℓ counts net photons emitted by the electron into the selected mode, assuming a nearly fixed photon frequency and a small relative energy change. For purely emissive transfers it is the emitted photon count; with emission and absorption it is the signed exchange count. It is not the total photon occupation of the cavity. In an optical-pump geometry the electron energy transfer must instead include the pump-photon exchange consistently with the Hamiltonian and chosen frame. Neglecting the accumulated mean energy loss relative to the electron energy requires
∣ ℓ ∣ η r e c ≪ 1. (152) \lvert\ell\rvert\eta_{\mathrm{rec}} \ll1. \tag*{(152)} ∣ ℓ ∣ η rec ≪ 1. (152)
This condition is stronger than requiring η r e c ≪ 1 \eta_{\mathrm{rec}} \ll1 η rec ≪ 1 alone. For a distribution of exchange orders, one must also control the appreciably populated tails and the induced width: cancellation of the mean signed exchange does not imply negligible recoil fluctuations or negligible electron–field correlations.
The Bloch–Nordsieck analogy is that an arbitrarily large photon population can accompany a small total radiative energy and momentum transfer. In the infrared spectrum, d N ∝ d ω / ω dN \propto d\omega/\omega d N ∝ d ω / ω , whereas d E = ℏ ω d N ∝ d ω dE = \hbar\omega dN \propto d\omega d E = ℏ ω d N ∝ d ω . Photon number may therefore diverge without an energy divergence [32 ] . Neglecting the finite accumulated transfer is an additional dynamical approximation. Radiation from a prescribed accelerated trajectory remains possible even when its radiation reaction is negligible.
Bandwidth, saturation and the scope of the approximation
For FEL resonance, the relevant comparison is more restrictive than the total relative energy criterion above. In conventional high-gain scaling, the electron energy acceptance is of order ρ 1 E e \rho_1 E_e ρ 1 E e . A negligible change of resonance due to the accumulated mean exchange requires, as a scale estimate,
∣ ℓ ∣ η r e c ρ 1 ≪ 1. (153) \frac{\lvert\ell\rvert\eta_{\mathrm{rec}}}{\rho_1} \ll1. \tag*{(153)} ρ 1 ∣ ℓ ∣ η rec ≪ 1. (153)
Order-unity coefficients depend on the adopted bandwidth and detuning conventions. This mean-shift criterion supplements the local phase condition and the distribution-width comparisons; it does not replace them. For example, a small mean exchange can coexist with an induced energy spread comparable with the resonance acceptance. The characteristic number of selected-mode photons transferred per electron when a fraction χ \chi χ of its energy is extracted is
ℓ s a t ∼ χ E e ℏ ω L = χ η r e c , ℓ s a t η r e c ∼ χ = O ( ρ 1 ) . (154) \ell_{\mathrm{sat}} \sim\frac{\chi E_e}{\hbar\omega_L} = \frac{\chi}{\eta_{\mathrm{rec}}}, \qquad\ell_{\mathrm{sat}}\eta_{\mathrm{rec}} \sim\chi= O(\rho_1). \tag*{(154)} ℓ sat ∼ ℏ ω L χ E e = η rec χ , ℓ sat η rec ∼ χ = O ( ρ 1 ) . (154)
This is an energy-balance estimate for the average net transfer, not a statement that each electron emits the same number of photons. For radiation power P r a d P_{\mathrm{rad}} P rad and electron beam power P E P_E P E , χ = P r a d / P E \chi= P_{\mathrm{rad}}/P_E χ = P rad / P E . The booklet convention used above gives χ s a t = 2 ρ 1 \chi_{\mathrm{sat}} = 2\rho_1 χ sat = 2 ρ 1 at P F , 1 = 2 ρ 1 P E P_{F,1} = 2\rho_1P_E P F , 1 = 2 ρ 1 P E ; the frequently used estimate χ s a t ∼ ρ 1 \chi_{\mathrm{sat}} \sim\rho_1 χ sat ∼ ρ 1 has the same parametric content. Thus the accumulated energy loss at saturation is small relative to the beam energy when ρ 1 ≪ 1 \rho_1 \ll1 ρ 1 ≪ 1 , yet is of the order of the FEL resonance acceptance. A fixed-trajectory approximation that suppresses this back-action cannot describe the approach to saturation self-consistently. Classical FEL theory retains the continuous evolution of electron energy, phase, bunching and field amplitude, while it can neglect the discreteness of the elementary quantum transfers. The distinction is particularly clear using the parameter already defined in the conclusions,
ρ ˉ = ρ 1 η r e c , ℓ s a t = O ( ρ ˉ ) , ℓ s a t η r e c ρ 1 = O ( 1 ) . (155) \bar{\rho} = \frac{\rho_1}{\eta_{\mathrm{rec}}}, \qquad\ell_{\mathrm{sat}} = O(\bar{\rho}), \qquad\frac{\ell_{\mathrm{sat}}\eta_{\mathrm{rec}}}{\rho_1} = O(1). \tag*{(155)} ρ ˉ = η rec ρ 1 , ℓ sat = O ( ρ ˉ ) , ρ 1 ℓ sat η rec = O ( 1 ) . (155)
When ρ ˉ ≫ 1 \bar{\rho} \gg1 ρ ˉ ≫ 1 , many elementary transfers fit within the characteristic extraction scale and a continuous classical description can be appropriate. Their cumulative effect nevertheless remains essential. When ρ ˉ \bar{\rho} ρ ˉ approaches unity or below, recoil discreteness can become dynamically important, subject to the beam preparation, coherence and diffusion limitations discussed in this article.
Finally, resonant FEL photons occupy a finite frequency band and their emission involves stimulated exchange and collective feedback. Negligible accumulated recoil therefore does not establish an infrared soft-photon limit. The precise conclusion is that a Bloch–Nordsieck-type no-recoil treatment is justified only when the total radiation-induced change in electron motion is negligible on the scales and over the duration relevant to the calculation. The inclusive infrared cancellation theorem is broader and does not require that total energy transfer vanish. This interpretation reconciles the usefulness of the prescribed-current reference with the necessity of retaining electron back-action in a working FEL.
Section 7 of Ref. [27 ] establishes the historical distinction between joint exchange-ladder coherence, recoil-induced statistical corrections and experimentally measured radiation coherence. Its warning about realistic beam distributions and longitudinal modes remains central to the observability problem. The prescribed-current source is gainless in its fixed-trajectory stimulated balance; a self-consistent classical electron trajectory can nevertheless yield gain with coherent conditional radiation.
The minimal-coupling and interaction-picture derivations identify detuning, recoil and photon exchange within one consistent dynamical framework. The prescribed-current solution supplies a coherent Poisson reference, whereas a quantum electron can become correlated with the field. The Raman–Nath equation makes recoil phases and their perturbative correction explicit, provided the populated ladder and its validity range are specified.
The physical control is the neighboring-state mismatch. Small ϵ \epsilon ϵ alone is insufficient if significantly occupied recoil orders accumulate large phases. A second limitation arises from the electron distribution: field-induced modulation and energy spread generate phase mixing in low gain as well as in collective amplification. Beam convolution supplies an initial response but does not replace self-consistent electron–field evolution.
The Wigner treatment defines the Moyal bracket and derives the finite momentum translation from the sinusoidal interaction. It distinguishes recoil populations from coherences responsible for spatial bunching. Smoothness on the recoil scale, accumulated recoil phase and beam dephasing are complementary criteria, rather than interchangeable measures of quantumness.
The oscillator comparison then embeds single-pass dynamics in a fresh-electron channel and cavity loss. Kling and Giese provide a recoil-resolved realization with many effective optical-undulator periods despite its short laboratory length and, in suitable nonlinear operating regions, sub-Poissonian photon statistics. This complements the perturbative Raman–Nath regime. The appendices distinguish low-gain quantum passages from high-gain SASE and explain how discrete-electron noise seeds collective amplification. They supply a dynamical comparison and a statistical baseline rather than independent evidence of observable quantum recoil.
Three competing effects must be compared quantitatively before interpreting a small quantum signature: stimulated exchange with the amplified selected mode, spontaneous emission into other radiation modes, and the self-consistent FEL-induced electron distribution change. The first and third are descriptions of the same selected-mode back-action in different regimes; their variances must not be added as independent sources. Spontaneous emission into unobserved modes is a separate contribution. Small initial energy spread, low gain per passage and small signal are not, individually, sufficient guarantees of observability. Bunch charge, duration and transverse size must also be specified: their collective Coulomb response can change gain and generate phase-dependent energy modulation. In the approximate high-gain equation of Ref. [37 ] , space charge and recoil enter through additive squared parameters, so intensity evolution alone does not disentangle them. The reported entrance spread and acceptance are compared in Eq. (139) ; Appendix § 1.14 separately analyzes correlated Coulomb modulation and the approximate high-gain space-charge/recoil degeneracy.
For the KG oscillator the stimulated contribution must be calculated with the stored cavity radiation, including photons retained from earlier passages. For a sharp upper-state electron momentum at resonance, the weak-transition probability is
P e m ( n ) ≃ ( g T ) 2 ( n + 1 ) = ( g T ) 2 + ( g T ) 2 n . (156) P_{\mathrm{em}}(n) \simeq(gT)^2(n+1) = (gT)^2 + (gT)^2n. \tag*{(156)} P em ( n ) ≃ ( g T ) 2 ( n + 1 ) = ( g T ) 2 + ( g T ) 2 n . (156)
The first term is spontaneous emission into the selected empty mode and the second is stimulated emission. To leading order the corresponding stimulated momentum variance is
σ p , s t i m 2 ≃ q 2 ( g T ) 2 n ‾ c a v , σ p , i n d 2 ≃ q 2 ( g T ) 2 ( n ‾ c a v + 1 ) . (157) \sigma_{p,\mathrm{stim}}^2 \simeq q^2(gT)^2\overline{n}_{\mathrm{cav}}, \qquad\sigma_{p,\mathrm{ind}}^2 \simeq q^2(gT)^2(\overline{n}_{\mathrm{cav}}+1). \tag*{(157)} σ p , stim 2 ≃ q 2 ( g T ) 2 n cav , σ p , ind 2 ≃ q 2 ( g T ) 2 ( n cav + 1 ) . (157)
The mean n ‾ c a v \overline{n}_{\mathrm{cav}} n cav must be obtained from pumping and cavity loss, not replaced by a vacuum input on every passage. More generally, in the resonant two-state model,
P ‾ e m = ∑ n P n c a v sin 2 ( g T n + 1 ) , σ p , i n d 2 = q 2 P ‾ e m ( 1 − P ‾ e m ) . (158) \overline{P}_{\mathrm{em}} = \sum_n P_n^{\mathrm{cav}} \sin^2\left(gT\sqrt{n+1}\right), \qquad\sigma_{p,\mathrm{ind}}^2 = q^2\overline{P}_{\mathrm{em}}(1-\overline{P}_{\mathrm{em}}). \tag*{(158)} P em = n ∑ P n cav sin 2 ( g T n + 1 ) , σ p , ind 2 = q 2 P em ( 1 − P em ) . (158)
The variance formula assumes an initially sharp electron momentum; finite-width inputs require the joint momentum–transition average and covariance of Eq. (89) . Photon-counting probabilities suffice for this ideal upper-state input, whereas a general electron preparation may also require coherence information. Let ξ = g T n ‾ c a v \xi= gT\sqrt{\overline{n}_{\mathrm{cav}}} ξ = g T n cav . Comparing the stimulated rms spread with the finite-time momentum acceptance discussed in Section XII gives the scale estimate
σ p , s t i m q / ( ω r T ) ∼ ξ ω r T . (159) \frac{\sigma_{p,\mathrm{stim}}}{q/(\omega_r T)} \sim\xi\omega_r T. \tag*{(159)} q / ( ω r T ) σ p , stim ∼ ξ ω r T . (159)
Consequently, ξ ≪ 1 \xi\ll1 ξ ≪ 1 need not make the induced spread negligible compared with a narrow recoil-selective acceptance when ω r T ≫ 1 \omega_r T \gg1 ω r T ≫ 1 . This comparison does not itself prove that resonance is destroyed: the ideal two-branch variance includes the intended recoil transition, not merely incoherent broadening within a branch. Fresh electrons prevent accumulation of the outgoing spread in the next bunch, but the stored field still controls each new passage. Branch widths, additional recoil orders and coherence must be checked at the actual intracavity intensity. Increasing that intensity also increases g n + 1 / ω r g\sqrt{n+1}/\omega_r g n + 1 / ω r , eventually invalidating the two-state approximation. Low gain therefore does not mean absence of stimulated electron back-action.
For high-gain SASE, even an initially cold beam accumulates uncorrelated spread from spontaneous undulator radiation. Define D γ = d σ γ 2 / d z D_\gamma= d\sigma_\gamma^2/dz D γ = d σ γ 2 / d z , with σ γ \sigma_\gamma σ γ the rms spread of the Lorentz factor. In an ideal planar magnetic undulator, with peak deflection parameter K K K , the standard diffusion expression is [58 ]
D γ = 7 15 r e λ ‾ C γ 4 k u 3 K 2 F ( K ) , F ( K ) = 1.20 K + 1 1 + 1.33 K + 0.40 K 2 . (160) D_\gamma= \frac{7}{15}r_e\overline{\lambda}_C\gamma^4 k_u^3 K^2 F(K), \qquad F(K) = 1.20K + \frac{1}{1+1.33K+0.40K^2}. \tag*{(160)} D γ = 15 7 r e λ C γ 4 k u 3 K 2 F ( K ) , F ( K ) = 1.20 K + 1 + 1.33 K + 0.40 K 2 1 . (160)
Here r e r_e r e is the classical electron radius, λ ‾ C = ℏ / ( m c ) \overline{\lambda}_C = \hbar/(mc) λ C = ℏ/ ( m c ) , and k u = 2 π / λ u k_u = 2\pi/\lambda_u k u = 2 π / λ u ; D γ D_\gamma D γ has units m − 1 \mathrm{m}^{-1} m − 1 . This uncorrelated diffusion has been measured at the European XFEL. The formula is for a magnetic undulator, not a direct substitution rule for the KG optical pump. Let Δ γ r e c = ℏ ω L / ( m c 2 ) \Delta\gamma_{\mathrm{rec}} = \hbar\omega_L/(mc^2) Δ γ rec = ℏ ω L / ( m c 2 ) be the laboratory energy change associated with one selected-mode radiation photon in the static-undulator case. The accumulated comparison is
σ γ , s p 2 ( L ) = ∫ 0 L D γ ( z ) d z , R d i f f ( L ) = σ γ , s p 2 ( L ) ( Δ γ r e c ) 2 . (161) \sigma_{\gamma,\mathrm{sp}}^{2}(L)=\int_{0}^{L}D_{\gamma}(z)\,\mathrm{d}z,\qquad\mathcal{R}_{\mathrm{diff}}(L)=\frac{\sigma_{\gamma,\mathrm{sp}}^{2}(L)}{(\Delta\gamma_{\mathrm{rec}})^{2}}. \tag*{(161)} σ γ , sp 2 ( L ) = ∫ 0 L D γ ( z ) d z , R diff ( L ) = ( Δ γ rec ) 2 σ γ , sp 2 ( L ) . (161)
For constant parameters and fundamental planar resonance,
R d i f f = 7 π 30 α f s N u K 2 F ( K ) ( 1 + K 2 2 ) 2 , L = N u λ u , (162) \mathcal{R}_{\mathrm{diff}}=\frac{7\pi}{30}\alpha_{\mathrm{fs}}N_{u}K^{2}F(K)\left(1+\frac{K^{2}}{2}\right)^{2},\qquad L=N_{u}\lambda_{u}, \tag*{(162)} R diff = 30 7 π α fs N u K 2 F ( K ) ( 1 + 2 K 2 ) 2 , L = N u λ u , (162)
where α f s = r e / λ ˉ C \alpha_{\mathrm{fs}}=r_{e}/\bar{\lambda}_{C} α fs = r e / λ ˉ C . For K = 1 K=1 K = 1 and N u = 1000 N_{u}=1000 N u = 1000 this gives R d i f f ≃ 18.9 \mathcal{R}_{\mathrm{diff}}\simeq18.9 R diff ≃ 18.9 , or an rms diffusion spread about 4.3 selected-mode recoils. This calculated illustration shows why a small relative beam spread can exceed the microscopic recoil scale.
When the recoil term in a normalized high-gain dispersion relation is of order [ Δ γ r e c / ( ρ γ ) ] 2 [\Delta\gamma_{\mathrm{rec}}/(\rho\gamma)]^{2} [ Δ γ rec / ( ρ γ ) ] 2 , while a narrow-distribution spread correction is of order [ σ γ / ( ρ γ ) ] 2 [\sigma_{\gamma}/(\rho\gamma)]^{2} [ σ γ / ( ρ γ ) ] 2 , their scale ratio is Eq. (161) . Exact coefficients and even the applicability of a variance expansion depend on the particular gain equation and distribution. Diffusion small compared with the ordinary FEL bandwidth can nevertheless exceed the small quantum correction being sought. The comparison must cover the amplification length required to observe the effect. It is not sufficient to characterize the entrance beam alone. In a drift–diffusion approximation the spontaneous contribution enters as ( D γ / 2 ) ∂ γ 2 f (D_{\gamma}/2)\partial_{\gamma}^{2}f ( D γ /2 ) ∂ γ 2 f in the longitudinal kinetic evolution, together with mean radiative energy loss. For rare large photon jumps an emission jump operator is required instead; the full spontaneous spectrum determines its moments [59 ] .
The stimulated high-gain broadening is quantified by the design booklet’s Eqs. (3.11)–(3.12) [60 ] . Denoting relative rms spreads by σ i \sigma_i σ i and σ ε \sigma_{\varepsilon} σ ε , those formulas read
σ i ( z ) ≃ 3 C A ( z ) 1 + 9 B [ A ( z ) − 1 ] , C = 1 2 ρ 1 P 0 P E , B ≃ 1.24 P 0 9 P F , 1 , σ ε 2 ( z ) = σ ε , 0 2 + σ i 2 ( z ) , σ i , F ≃ 1.6 ρ 1 . (163) \begin{aligned}
\sigma_i(z)&\simeq3C\sqrt{\frac{A(z)}{1+9B[A(z)-1]}},\qquad C=\frac{1}{2}\sqrt{\frac{\rho_1P_0}{P_E}},\\
B&\simeq\frac{1.24P_0}{9P_{F,1}},\qquad\sigma_{\varepsilon}^{2}(z)=\sigma_{\varepsilon,0}^{2}+\sigma_i^{2}(z),\qquad\sigma_{i,F}\simeq1.6\rho_1. \tag*{(163)}
\end{aligned} σ i ( z ) B ≃ 3 C 1 + 9 B [ A ( z ) − 1 ] A ( z ) , C = 2 1 P E ρ 1 P 0 , ≃ 9 P F , 1 1.24 P 0 , σ ε 2 ( z ) = σ ε , 0 2 + σ i 2 ( z ) , σ i , F ≃ 1.6 ρ 1 . (163)
Here P E P_E P E is the beam power, P 0 P_0 P 0 the input radiation power, ρ 1 \rho_1 ρ 1 the fundamental Pierce parameter and P F , 1 = 2 ρ 1 P E P_{F,1}=2\rho_1P_E P F , 1 = 2 ρ 1 P E the booklet’s saturation power. Its linear growth function is
A ( z ) = 1 9 [ 3 + 2 cosh ( z L g , 1 ) + 4 cos ( 3 z 2 L g , 1 ) cosh ( z 2 L g , 1 ) ] . (164) A(z)=\frac{1}{9}\left[3+2\cosh\left(\frac{z}{L_{g,1}}\right)+4\cos\left(\frac{\sqrt{3}z}{2L_{g,1}}\right)\cosh\left(\frac{z}{2L_{g,1}}\right)\right]. \tag*{(164)} A ( z ) = 9 1 [ 3 + 2 cosh ( L g , 1 z ) + 4 cos ( 2 L g , 1 3 z ) cosh ( 2 L g , 1 z ) ] . (164)
These are approximate classical design formulas, not an exact initial-value solution for arbitrary quantum preparations. In the exponential interval before saturation, using P ( z ) ≃ P 0 A ( z ) P(z)\simeq P_0A(z) P ( z ) ≃ P 0 A ( z ) and neglecting the slowly varying fitting factor in the booklet’s power formula, they imply
σ i 2 ( z ) ≃ 9 4 ρ 1 P ( z ) P E . (165) \sigma_i^{2}(z)\simeq\frac{9}{4}\rho_1\frac{P(z)}{P_E}. \tag*{(165)} σ i 2 ( z ) ≃ 4 9 ρ 1 P E P ( z ) . (165)
An effective startup power may be used for a SASE scaling estimate, but a fluctuating multimode SASE field requires its own statistical evolution rather than identification with a deterministic seed.
Define η r e c = ℏ ω L / E e \eta_{\mathrm{rec}}=\hbar\omega_L/E_e η rec = ℏ ω L / E e and ρ ˉ = ρ 1 / η r e c \bar{\rho}=\rho_1/\eta_{\mathrm{rec}} ρ ˉ = ρ 1 / η rec , where E e = γ m c 2 E_e=\gamma mc^2 E e = γ m c 2 . The stimulated variance relative to the recoil-squared scale is then
R F E L ( z ) = σ i 2 ( z ) η r e c 2 ≃ 9 4 ρ ˉ 2 P ( z ) ρ 1 P E . (166) \mathcal{R}_{\mathrm{FEL}}(z)=\frac{\sigma_i^{2}(z)}{\eta_{\mathrm{rec}}^{2}}\simeq\frac{9}{4}\bar{\rho}^{2}\frac{P(z)}{\rho_1P_E}. \tag*{(166)} R FEL ( z ) = η rec 2 σ i 2 ( z ) ≃ 4 9 ρ ˉ 2 ρ 1 P E P ( z ) . (166)
For a small additive recoil correction of order 1 / ρ ˉ 2 1/\bar{\rho}^{2} 1/ ρ ˉ 2 , stimulated spread can exceed this scale when
P ( z ) P F , 1 ≫ 2 9 ρ ˉ 2 . (167) \frac{P(z)}{P_{F,1}}\gg\frac{2}{9\bar{\rho}^{2}}. \tag*{(167)} P F , 1 P ( z ) ≫ 9 ρ ˉ 2 2 . (167)
For example, at ρ ˉ = 100 \bar{\rho}=100 ρ ˉ = 100 , equality occurs at P / P F , 1 ≃ 2.2 × 10 − 5 P/P_{F,1} \simeq2.2 \times10^{-5} P / P F , 1 ≃ 2.2 × 1 0 − 5 , well below saturation. The practical obstacle is therefore not only externally supplied energy spread: the amplification needed to reveal a small correction can itself generate a larger change in the electron distribution.
This comparison must retain its dynamical meaning. FEL-induced spread is correlated with electron phase, bunching and field growth. It cannot generally be substituted for a fixed Gaussian entrance spread or an independent diffusion coefficient. In particular, inserting Eq. (163) into a cold-beam cubic gain equation is a scale estimate, not a self-consistent derivation. The evolving inhomogeneous response can require a kinetic or integral formulation. Likewise, a variance exceeding the recoil-squared scale does not alone prove that every quantum observable is erased: one must compare the actual observable calculated with stimulated evolution, spontaneous diffusion and the compatible beam preparation. The classical booklet estimates support a serious limitation for small perturbative quantum corrections; they do not establish a universal exclusion of the recoil-resolved regime ρ ˉ ≲ 1 \bar{\rho} \lesssim1 ρ ˉ ≲ 1 , where a quantum self-consistent treatment replaces those classical estimates.
A meaningful experimental claim must show that its specified recoil or correlation observable survives averaging, technical fluctuations, loss and mode selection under achievable beam and undulator conditions. Neither startup alone nor narrowing relative to another broad distribution establishes that claim.
This requirement leaves a constructive experimental direction. Carmesin and collaborators identify how microscopic quantum structure and averaged gain can differ; Kapitza–Dirac exchange offers a controllable momentum-space interaction; and Walther’s micromaser research supplies a precedent for particle preparation and cavity readout. A dedicated few-electron, low-occupation experiment with deflection–photon coincidences can bring these perspectives together. The aim is to test specified joint observables under characterized conditions, while establishing the coupling and coherence needed for an eventual quantum FEL experiment.
Recoil observability therefore depends on accumulated transition phases, electron and optical preparation, and the acceptance of the measured interaction. Neither spontaneous startup nor excess photon variance alone establishes a nonclassical FEL signature.
Acknowledgments and AI-use disclosure
During the preparation of this article, the authors used OpenAI’s ChatGPT, including the Codex assistant in Work mode, to assist with calculations, checks of derivations, literature searches, numerical cross-checks, and revision of the English text. The model versions used at different stages of the work were not systematically recorded; the final revision was checked using GPT-5.6 Sol.
The authors conceived and directed the study, drawing on their earlier work on the quantum theory of free-electron lasers and on subsequent contributions by other researchers. AI-assisted calculations were used to check analytical results and to compare calculations performed with Mathcad and Mathematica. AI tools also assisted in locating relevant publications and in examining concise derivations in the literature. The cited publications were consulted by the authors.
All scientific interpretations, approximations, analytical and numerical results, and the final wording were reviewed by the authors, who take full responsibility for the content of the article.
Low-gain quantum FEL dynamics and high-gain SASE
Different dynamical problems
The low-gain quantum description resolves joint momentum–photon exchange during one passage. An incoming electron state and optical state specify a unitary transition kernel; electron tracing and entrance averaging then give observable probabilities. Low gain does not imply a single-electron beam: an independent-passage reduction can average over many electrons, while a bunch with appreciable common-field correlations requires its full channel. In an oscillator, fresh-electron pumping and losses evolve the stored field over successive passages.
High-gain SASE is a many-electron traveling-wave initial-value problem. A finite beam supplies a stochastic source; the common field changes electron phases and energies, which change the source in turn. Slippage, longitudinal modes and the evolving distribution matter. A sum of independent emission probabilities cannot replace this feedback. Recoil can be included in a self-consistent many-electron ladder or density-matrix description [9 , 16 ] ; its classical limit retains collective dynamics while smoothing elementary transfers.
Table 2.
A conditional low-gain Poisson law does not establish SASE statistics. Classical high-gain bunching and growth do not alone identify recoil. Both regimes require propagation of compatible entrance information to the same measured observable.
Space charge, energy modulation and the high-gain response
The following discussion concerns a collective beam and its high-gain response. It supplies a broader observability comparison; it is not a quantitative correction to the low-gain Kling–Giese two-state cavity model. Charge-density and plasma-frequency quantities were defined in Eqs. (138) –(141) . The distinction between correlated Coulomb chirp, uncorrelated entrance spread and spontaneous diffusion is essential when using an intensity-based gain equation.
Space charge redistributes energy and phase, and can generate a correlated longitudinal chirp. It is not a photon-emission diffusion coefficient. A simple estimate makes the dependence on geometry explicit: for a cold uniform ellipsoid of laboratory density n e n_e n e , let D z D_z D z be the longitudinal depolarization factor of its rest-frame shape. The linear internal electric force in laboratory coordinates is F z ≃ e 2 n e D z z / ϵ 0 F_z \simeq e^2 n_e D_z z/\epsilon_0 F z ≃ e 2 n e D z z / ϵ 0 . Over a short interval with negligible expansion,
σ γ , S C ∼ β e 2 n e D z T l a b σ z ϵ 0 m c , σ p , S C ∼ e 2 n e D z T l a b σ z ϵ 0 . (168) \sigma_{\gamma,\mathrm{SC}} \sim\frac{\beta e^2 n_e D_z T_{\mathrm{lab}}\sigma_z}{\epsilon_0 m c}, \qquad\sigma_{p,\mathrm{SC}} \sim\frac{e^2 n_e D_z T_{\mathrm{lab}}\sigma_z}{\epsilon_0}. \tag*{(168)} σ γ , SC ∼ ϵ 0 m c β e 2 n e D z T lab σ z , σ p , SC ∼ ϵ 0 e 2 n e D z T lab σ z . (168)
These expressions are the rms width of the induced linear chirp, not an independent variance to add blindly to an already correlated distribution. The estimate follows from Δ p z = F z T l a b \Delta p_z = F_z T_{\mathrm{lab}} Δ p z = F z T lab and Δ γ ≃ β Δ p z / ( m c ) \Delta\gamma\simeq\beta\Delta p_z/(mc) Δ γ ≃ β Δ p z / ( m c ) . It requires Ω p T l a b ≪ 1 \Omega_pT_{\mathrm{lab}} \ll1 Ω p T lab ≪ 1 for the frozen-shape approximation and excludes acceleration, conducting boundaries and external compensation. The uniform-ellipsoid density is not the Gaussian central density of Eq. (138) . The factor 0 < D z < 1 0 < D_z < 1 0 < D z < 1 depends on the transformed aspect ratio; a Gaussian bunch or a beam in conducting boundaries requires its actual Poisson field. Initial correlations, focusing and expansion must be propagated for a quantitative result. The induced width must be compared with the recoil-equivalent energy width and resonance acceptance, including slice-dependent detuning; a removable mean chirp and uncorrelated slice spread affect observations differently.
The connection with the collective analysis of Ref. [37 ] is particularly direct. In the notation and approximate one-dimensional normalization of Eq. (37) of Ref. [37 ] ,
a ( 3 ) + 2 i ϕ ~ a ′ ′ + ( μ ~ Q 2 + μ ~ q 2 − ϕ ~ 2 ) a ′ = i 3 3 a , μ ~ q = ℏ ω L γ γ m c 2 = 1 ρ . (169) a^{(3)} + 2i\widetilde{\phi}a'' + \left(\widetilde{\mu}_Q^2+\widetilde{\mu}_q^2-\widetilde{\phi}^2\right)a' = \frac{i}{3\sqrt{3}}a,\qquad\widetilde{\mu}_q = \frac{\hbar\omega_L}{\gamma\gamma mc^2} = \frac{1}{\rho}. \tag*{(169)} a ( 3 ) + 2 i ϕ a ′′ + ( μ Q 2 + μ q 2 − ϕ 2 ) a ′ = 3 3 i a , μ q = γ γ m c 2 ℏ ω L = ρ 1 . (169)
Primes are derivatives with respect to z ~ = z / L g \widetilde{z}=z/L_g z = z / L g , L g = λ u / ( 4 π 3 ρ ) L_g=\lambda_u/(4\pi\sqrt{3}\rho) L g = λ u / ( 4 π 3 ρ ) , and ϕ ~ = ( ω 0 − ω ) / ( 2 3 ρ ω 0 ) \widetilde{\phi}=(\omega_0-\omega)/(2\sqrt{3}\rho\omega_0) ϕ = ( ω 0 − ω ) / ( 2 3 ρ ω 0 ) . The coefficient μ ~ Q \widetilde{\mu}_Q μ Q is the space-charge parameter of the paper’s high-gain model; its numerical value must be retained when converting a physical plasma phase into the reported coefficient. This is the reference’s phenomenological recoil extension of a cold-beam equation, not a derivation of the full many-electron quantum hierarchy or a formula to transplant unchanged into the optical cavity model.
Within that stated equation, the two effects enter only through M 2 = μ ~ Q 2 + μ ~ q 2 M^2=\widetilde{\mu}_Q^2+\widetilde{\mu}_q^2 M 2 = μ Q 2 + μ q 2 . An intensity evolution alone therefore cannot identify their separate contributions. A comparison must independently determine the charge-density response and vary charge, current or geometry under controlled beam conditions. Increasing recoil or reducing entrance energy spread does not remove this degeneracy. On the same coefficient scale the ratio is μ ~ Q 2 / μ ~ q 2 \widetilde{\mu}_Q^2/\widetilde{\mu}_q^2 μ Q 2 / μ q 2 ; collective Coulomb response can exceed the proposed recoil correction even while both perturb ordinary gain only weakly.
For the homogeneous model with a ( 0 ) a(0) a ( 0 ) prescribed and a ′ ( 0 ) = a ′ ′ ( 0 ) = 0 a'(0)=a''(0)=0 a ′ ( 0 ) = a ′′ ( 0 ) = 0 , the equivalent memory equation is
d a d z ~ = i 3 3 ∫ 0 z ~ e − i ϕ ~ s sin ( M s ) M a ( z ~ − s ) d s , sin ( M s ) M ⟶ s ( M → 0 ) . (170) \frac{da}{d\widetilde{z}}=\frac{i}{3\sqrt{3}}\int_0^{\widetilde{z}}e^{-i\widetilde{\phi}s}\frac{\sin(Ms)}{M}a(\widetilde{z}-s)\,ds,\qquad\frac{\sin(Ms)}{M}\longrightarrow s\quad(M\to0). \tag*{(170)} d z d a = 3 3 i ∫ 0 z e − i ϕ s M sin ( M s ) a ( z − s ) d s , M sin ( M s ) ⟶ s ( M → 0 ) . (170)
It follows by factoring the operator acting on a ′ a' a ′ as ( ∂ z + i ϕ ~ ) 2 + M 2 (\partial_z+i\tilde{\phi})^2+M^2 ( ∂ z + i ϕ ~ ) 2 + M 2 . A distribution of detunings modifies the kernel; a Gaussian static distribution multiplies its phase factor by e − σ ϕ 2 s 2 / 2 e^{-\sigma_\phi^2s^2/2} e − σ ϕ 2 s 2 /2 . Inhomogeneous or evolving beams generally retain a Volterra response rather than a universal constant-coefficient cubic. Ref. [37 ] stresses both the similarity of gain dilution and the heuristic limitations of its quantum extension. A measured photon-statistical or electron–photon correlation observable must therefore be calculated beyond an intensity-only comparison if it is to distinguish quantum information from collective classical forces.
Shot noise and the early stages of FEL growth
An exactly smooth deterministic Maxwell–Lorentz or Vlasov beam with a uniform phase distribution and zero optical field can remain a symmetry-preserving solution. A real beam contains a finite number of electrons. Their phases do not form a perfectly smooth distribution in a single realization. This discreteness gives electron shot noise. For N N N independent phases θ j \theta_j θ j uniformly distributed over [ 0 , 2 π ) [0,2\pi) [ 0 , 2 π ) ,
b N = 1 N ∑ j = 1 N e − i θ j , ⟨ b N ⟩ = 0 , ⟨ ∣ b N ∣ 2 ⟩ = 1 N . (171) b_N=\frac{1}{N}\sum_{j=1}^{N}e^{-i\theta_j},\qquad\langle b_N\rangle=0,\qquad\langle\lvert b_N\rvert^2\rangle=\frac{1}{N}. \tag*{(171)} b N = N 1 j = 1 ∑ N e − i θ j , ⟨ b N ⟩ = 0 , ⟨∣ b N ∣ 2 ⟩ = N 1 . (171)
The variance follows by separating the diagonal j = k j=k j = k terms from off-diagonal terms whose independent phase averages vanish. Thus the rms microscopic bunching is N − 1 / 2 N^{-1/2} N − 1/2 with a random phase, rather than a nonzero coherent ensemble mean.
The field couples to bunching and changes electron energies and phases. A random seed can excite an unstable eigenmode of the coupled system. During a linear high-gain interval a component has the schematic form
a ( τ ) ∼ a n o i s e e λ τ , b ( τ ) ∼ b n o i s e e λ τ , Re λ > 0. (172) a(\tau)\sim a_{\mathrm{noise}}e^{\lambda\tau},\qquad b(\tau)\sim b_{\mathrm{noise}}e^{\lambda\tau},\qquad\operatorname{Re}\lambda>0. \tag*{(172)} a ( τ ) ∼ a noise e λ τ , b ( τ ) ∼ b noise e λ τ , Re λ > 0. (172)
Shot noise supplies the seed; the instability of the coupled electron–field system supplies gain. For a nonuniform beam or a time-dependent kernel the growth law need not reduce to a single constant exponential. The expression is a local linear-mode illustration, not a universal cubic reduction.
If A s a t A_{\mathrm{sat}} A sat marks the end of the linear regime, then
τ s a t ∼ 1 Re λ log ( A s a t A n o i s e ) . (173) \tau_{\mathrm{sat}}\sim\frac{1}{\operatorname{Re}\lambda}\log\left(\frac{A_{\mathrm{sat}}}{A_{\mathrm{noise}}}\right). \tag*{(173)} τ sat ∼ Re λ 1 log ( A noise A sat ) . (173)
For a fixed gain rate, seed strength shifts startup only logarithmically. Changing the number of electrons can also change current and gain, so the N − 1 / 2 N^{-1/2} N − 1/2 seed scaling must not be used while silently holding all beam dynamics invariant.
Three statements must be distinguished. A perfectly smooth deterministic zero-field beam does not create a preferred phase by itself. A discrete classical beam radiates and can seed collective amplification without quantizing the electromagnetic field. A fully quantum electron–field description contains vacuum fluctuations and the n + 1 n+1 n + 1 emission factor, so emission is possible into an empty mode. Observing startup alone cannot distinguish these descriptions.
Shot noise and recoil act at different levels. Electron discreteness changes the stochastic source; recoil changes elementary transitions and their phase dependence. A realistic calculation must include both without duplicating the same spontaneous feeding in two representations. The comparison should use an ensemble of classical noise realizations propagated through the same collective dynamics and detection model, not merely compare a quantum probability law with a smooth zero-field calculation.
Classical source and technical fluctuations can give super-Poissonian photon statistics. Within a semiclassical mixture of coherent fields, F = 1 + Var ( μ ) / ⟨ μ ⟩ ≥ 1 F=1+\operatorname{Var}(\mu)/\langle\mu\rangle\ge1 F = 1 + Var ( μ ) / ⟨ μ ⟩ ≥ 1 , where μ \mu μ is the random conditional mean. For a specified counting gate and collection of optical modes, statistically significant sub-Poissonian photon counts are a stronger nonclassical diagnostic than excess variance or startup. The measured counts must be related to the source through a calibrated detector response: independent losses move F F F toward unity, while background counts and electronic noise contribute their own means and variances. Detector saturation or dead time can artificially suppress fluctuations and must be excluded. If several modes are collected, the variance includes their individual variances and cross-covariances; the measured quantity must be compared with a prediction for that same collection. These conditions specify how a reported F < 1 F < 1 F < 1 is assessed, rather than assuming an unspecified correction for “background and multimode effects.” Electron–photon correlations and conditional distributions can provide additional discriminants. The shot-noise discussion supplies the classical statistical baseline; the phase-space study of [43 ] further illustrates why quantum structure need not survive observable averaging.
References