betaFeedback

WorksPhysics and Astronomy

Single-pass stochastic cooling for electron microscopy

Abstract

Scientific applications of free electrons, such as in electron microscopes and particle accelerators, are limited by the brightness of available electron sources. Accelerators can use phase space cooling to surpass these source limitations, whereas electron microscopes can only rely on particle filtering to increase beam coherence at the cost of reduced brightness. We show that strong electron-light coupling in nanophotonic structures makes it possible to achieve single-pass stochastic cooling inside an electron microscope for a lossless improvement of longitudinal emittance. This approach is laser-free and compatible with normal CW operation of electron microscopes. We derive a fully quantum mechanical model of stochastic cooling valid in the strong coupling regime, finding an analytic asymptotic formula for the optimal cooled energy spread σEf≃0.88(ℏω)2/3σE01/3σ_{E_f}\simeq0.88(\hbarω)^{2/3}σ_{E_0}^{1/3}, with ℏω\hbar ω the cooling photon energy and σE0σ_{E_0} the initial energy spread. Numerical simulations confirm that the energy spread reduction is reflected in an increase in the purity of the electron's quantum state and a reduction of the longitudinal phase space area. Our work paves a path towards even brighter beams for electron microscopy, promising higher image quality, enhanced spectral resolution, and a route to reaching fundamental quantum limits on electron beam phase space area.

Introduction

High-brightness electron beams underpin critical scientific instruments from large particle accelerators [1] to tabletop electron microscopes [2]. Electron microscopes combine tip emitters with spatial and energetic filters to achieve sub-Å spatial resolution and sub-eV energy widths [3, 4]. Ultrafast electron microscopes additionally use photoemission to confine electron pulses below one picosecond, typically with about one electron per pulse [5, 6, 7, 8, 9, 10]. Tip-based sources are among the closest we have to producing quantum-uncertainty-limited electron beams, yet are still orders of magnitude away from fundamental quantum limits [11, 12, 13]. Small energy spreads in particular are important both for direct spectroscopy as in electron energy loss spectroscopy and for minimizing chromatic aberrations in imaging. The energy width can be further reduced with monochromators which remove off-energy electrons [14, 15] or with active phase space manipulations such as THz modulation [16, 17, 18]. Both have disadvantages. The first reduces the usable current, while the second is limited by Liouville’s theorem and cannot improve the longitudinal emittance dictated by the electron source.

Phase space cooling techniques avoid both disadvantages by moving entropy from the electron beam into a different system. To our knowledge, no such scheme has been proposed or demonstrated in an electron microscope. In accelerators, on the other hand, techniques like stochastic cooling (SC) [19, 20, 21] are well-established. In transit-time optical stochastic cooling (OSC) [22, 23], spontaneous radiation emitted in a pickup is used to later remove energy fluctuations from the electron in a kicker. OSC was recently demonstrated at the Integrable Optics Test Accelerator with 100 MeV electron beams [24, 25]. OSC must generally be performed in a storage ring over many passes due to the large initial energy spread, limited spontaneous undulator radiation emission rates (about α=1/137\alpha= 1/137 photons per electron in the fundamental mode [26]), and incoherent kicks from other particles in the bunch.

In this Letter, we propose to extend stochastic cooling to electron microscopes. Unlike in accelerators, in microscopes the energy spread to be cooled is at the eV level. This corresponds to no more than tens of (mid-infrared) photons, requiring a fully quantum treatment of the electron and the radiation. Critically, electron microscopes can leverage nanophotonic structures enabling higher electron-light coupling efficiencies than are possible in undulators [27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42]. We show that high coupling efficiency makes single-pass cooling possible, making the scheme naturally compatible with electron microscope columns. This approach is also laser-free, making it compatible with standard CW operation of electron microscopes. We develop a fully quantum mechanical model of the stochastic cooling process in the language of free-electron quantum optics [43, 44, 45], extending earlier quantum treatments of OSC [46] to the strong-coupling regime. We derive optimal conditions for cooling yielding a minimal energy spread σEf≃0.88(ℏω)2/3σE01/3\sigma_{E_f} \simeq0.88(\hbar\omega)^{2/3}\sigma_{E_0}^{1/3}, with ℏω\hbar\omega the cooling photon energy and σE0\sigma_{E_0} the initial energy spread. With numerical tracking of the electron density matrix, we show that this reduced energy spread corresponds to true phase space cooling, reflected as an increase in the purity of the electron’s quantum state and a reduction of its longitudinal emittance. This paves a path towards bringing the already impressive brightness of electron microscopes closer to its fundamental quantum limits, which is of highest practical and fundamental interests.

Theory

OSC consists of three stages (Figure 1): (1) the “pickup” where the electron spontaneously emits radiation, (2) the drift during which the electron dephases from the light depending on its energy, and (3) the “kicker” in which the light emitted in the pickup modulates the electron energy. The pickup radiation encodes the arrival time of the electron in its phase, naturally enabling electron-light synchronization in the kicker. The drift is designed such that an electron with the desired reference energy reaches the kicker at the zero crossing of the pickup wave. In that case, faster (slower) electrons will meet the wave at a decelerating (accelerating) phase, leading to cooling of the total energy spread. (a), (b), and (c) show the relative electron-light overlap after the pickup, drift, and kicker, respectively.

Figure 1. Concept for single-pass stochastic cooling. In a pickup, the electron emits spontaneous radiation whose phase encodes the electron arrival time. After a dispersive drift, the electron meets the radiation in the kicker, where fast (slow) electrons end up in a decelerating (accelerating) phase, cooling the total energy spread. (a), (b), and (c) show the electron-light overlap after the pickup, drift, and kicker, respectively.

To model this three-stage process, we treat both the electron and the light quantum mechanically. The electron moves in the zz direction with average velocity v0v_0. The optical mode has frequency ω\omega and wavenumber in the zz direction kk. We assume that the optical mode satisfies the phase-matching condition ω=kv0\omega= kv_0. We write the electron momentum operator in the scaled form m^=p^/ℏk\hat{m} = \hat{p}/\hbar k, where p^\hat{p} is the momentum operator relative to the nominal momentum. We consider a single electron active in the system at a time. We estimate the validity of this assumption in the Supplementary Material (SM) [47]. Passage through the pickup, drift, or kicker can be modeled using the following scattering operator, which evolves the entangled electron-light state (see SM for derivation):

S^(g,κ)=e−iκm^2+(gb^a^†−g∗b^†a^),(1)\hat{S}(g,\kappa) = e^{-i\kappa\hat{m}^{2}+(g\hat{b}\hat{a}^{\dagger}-g^{*}\hat{b}^{\dagger}\hat{a})}, \tag*{(1)}

where κ=Lℏk22γ03mev0\kappa= \frac{L\hbar k^{2}}{2\gamma_{0}^{3}m_{e}v_{0}} is the normalized section length and g=−eLEz∗ℏωg = -\frac{eLE_{z}^{*}}{\hbar\omega} the vacuum coupling strength, with EzE_z the single mode electric field strength. b^=e−ikz^\hat{b} = e^{-ik\hat{z}} is a ladder operator which reduces the electron momentum by ℏk\hbar k, and a^\hat{a} is the annihilation operator of the optical mode. Taking the no-recoil approximation in the pickup and kicker, the three-stage OSC transformation can be represented by the combined scattering operator

S^T(g1,g2,κ)=S^(g2,0)S^(0,κ)S^(g1,0)(2)\hat{S}_{T}(g_{1},g_{2},\kappa)=\hat{S}(g_{2},0)\hat{S}(0,\kappa)\hat{S}(g_{1},0) \tag*{(2)}
=eg2b^a^†−H.c.e−iκm^2eg1b^a^†−H.c..(3)= e^{g_{2}\hat{b}\hat{a}^{\dagger}-\mathrm{H.c.}}e^{-i\kappa\hat{m}^{2}}e^{g_{1}\hat{b}\hat{a}^{\dagger}-\mathrm{H.c.}.} \tag*{(3)}

g1g_1 and g2g_2 are the complex coupling factors for the pickup and kicker, respectively, and κ\kappa is the scaled drift length between them. Recoil effects in the pickup and kicker are treated numerically in the SM. We find that most of the recoil effects can be compensated by slightly reducing the drift length κ\kappa for expected parameters.

To start, we consider an electron initially in a momentum eigenstate with relative momentum p0=ℏkm0p_0=\hbar k m_0 and the photons in the vacuum state. After the kicker, the electron will have emitted some average number of photons ⟨nph(m0)⟩\langle n_{\rm ph}(m_0)\rangle depending on its initial momentum m0m_0. We show in the SM that this is

⟨nph(m0)⟩=∣g1∣2+∣g2∣2+2∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)cos⁡(2κm0+ψ),(4)\langle n_{\rm ph}(m_0)\rangle= |g_1|^2+|g_2|^2 +2|g_1||g_2|e^{-2|g_1|^2\sin^2(\kappa)}\cos(2\kappa m_0+\psi), \tag*{(4)}

where ψ=Δϕg−κ−∣g1∣2sin⁡(2κ)\psi=\Delta\phi_g-\kappa-|g_1|^2\sin(2\kappa), with g1g2∗=∣g1∣∣g2∣eiΔϕgg_1g_2^{*}=|g_1||g_2|e^{i\Delta\phi_g}. Neglecting briefly the fluctuations of the emitted photon number, the final scaled momentum of the electron is simply mf(m0)=m0−⟨nph(m0)⟩m_f(m_0)=m_0-\langle n_{\rm ph}(m_0)\rangle. The lowest-order cooling condition is that the final momentum is independent of the initial momentum: dmfdm0∣m0=0=0\left.\frac{dm_f}{dm_0}\right|_{m_0=0}=0, or 4κ∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)sin⁡(ψ)=−14\kappa|g_1||g_2|e^{-2|g_1|^2\sin^2(\kappa)}\sin(\psi)=-1. All scaled parameters are positive, so we must choose sin⁡(ψ)<0\sin(\psi)<0. Furthermore, the second-order curvature in the final phase space, d2mfdm02∣m0=0\left.\frac{d^2m_f}{dm_0^2}\right|_{m_0=0}, vanishes for ψ=−π2\psi=-\frac{\pi}{2}. This choice constrains the relative phase of the couplings Δϕg\Delta\phi_g. This implies the lowest-order cooling condition

4κ∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)=1.(5)4\kappa|g_1||g_2|e^{-2|g_1|^2\sin^2(\kappa)}=1. \tag*{(5)}

To obtain more precise conditions for optimal cooling, we consider an electron whose initial momentum is distributed according to the distribution function f(m)=12πσm0e−m22σm02f(m)=\frac{1}{\sqrt{2\pi\sigma_{m_0}}}e^{-\frac{m^2}{2\sigma_{m_0}^{2}}}, with σm0=σE0/ℏω\sigma_{m_0}=\sigma_{E_0}/\hbar\omega the rms energy spread in units of the photon energy. The final energy spread in this case has two contributions: (1) electrons with different initial momentum emit a different number of photons on average, and (2) the number of emitted photons fluctuates even for definite initial momentum. We evaluate both contributions as a function of arbitrary parameters in the SM, including an exact expression for the final energy spread as a function of arbitrary ∣g1∣|g_1|, ∣g2∣|g_2|, ψ\psi, and κ\kappa. For ψ=−π/2\psi=-\pi/2 we find an exact optimal value for ∣g2∣|g_2|:

∣g2∣opt=4κ∣g1∣e−2κ2σmeff2(σm02+∣g1∣2sin⁡(κ)cos⁡(κ))1+2∣g1∣2χ,(6)|g_2|_{\rm opt}=\frac{4\kappa|g_1|e^{-2\kappa^2\sigma_{m_{\rm eff}}^2}\left(\sigma_{m_0}^2+|g_1|^2\sin(\kappa)\cos(\kappa)\right)} {1+2|g_1|^2\chi}, \tag*{(6)}

where χ=1−e−8κ2σmeff2+8∣g1∣2sin⁡4(κ)cos⁡(2κ−8∣g1∣2cos⁡(κ)sin⁡3(κ))\chi= 1-e^{-8\kappa^2\sigma_{m_{\rm eff}}^2+8|g_1|^2\frac{\sin^4(\kappa)}{\cos(2\kappa-8|g_1|^2\cos(\kappa)\sin^3(\kappa))}}, with \sigma_{m_{\rm eff}}^2=\sigma_{m_0}^2+|g_1|^2\sinc^2(\kappa) the effective energy spread after the pickup and drift. At this optimal value, the final energy spread takes the relatively simple form

\sigma_{m_f}^2=\sigma_{m_0}^2\left[1-4\kappa|g_1||g_2|e^{-2\kappa^2\sigma_{m_{\rm eff}}^2}\right] +|g_1|^2\left[1-4\kappa|g_1||g_2|e^{-2\kappa^2\sigma_{m_{\rm eff}}^2}\sinc(\kappa)\cos(\kappa)\right]. \tag*{(7)}

As we show in the SM, for κ≪1\kappa\ll1 and ∣g1∣≫1|g_1|\gg1, this permits an analytic optimal value of the drift length

κopt≃12σmeff(31+4∣g1∣2)1/6.(8)\kappa_{\mathrm{opt}}\simeq\frac{1}{2\sigma_{m_{\rm eff}}}\left(\frac{3}{1+4|g_1|^2}\right)^{1/6}. \tag*{(8)}

At this condition the energy spread is approximately

σmf2≃32/34×21/3σm02+∣g1∣2∣g1∣4/3,\sigma_{m_f}^2\simeq\frac{3^{2/3}}{4\times2^{1/3}}\frac{\sigma_{m_0}^2+|g_1|^2}{|g_1|^{4/3}},

which is minimized for ∣g1∣opt=2σm0|g_1|_{\mathrm{opt}}=\sqrt{2}\sigma_{m_0} at value σmf,opt≃35/38σm01/3\sigma_{m_f,\mathrm{opt}}\simeq\sqrt{\frac{3^{5/3}}{8}}\sigma_{m_0}^{1/3}. In eV, this means σEf,opt≃0.88(ℏω)2/3σE01/3\sigma_{E_f,\mathrm{opt}}\simeq0.88(\hbar\omega)^{2/3}\sigma_{E_0}^{1/3}. Lower photon energies are preferred for a given input energy spread, though this implies larger couplings ∣g1∣|g_1| and longer pickup and kicker sections.

Figure 2 gives an overview of the single-pass cooling performance using Eqn. (7). Panel (a) shows the final spread σmf\sigma_{m_f}, in units of the photon energy, as a function of the pickup coupling ∣g1∣|g_1| and drift length κ\kappa for σm0=3\sigma_{m_0}=3 – an initial rms energy spread of 3 photon energies. The white region coincides with σmf=σm0\sigma_{m_f}=\sigma_{m_0}, the blue represents cooling, and the red represents heating. The orange line shows the optimal drift length estimate according to Eqn. (8). The estimate agrees very well with the false color plot for all pickup couplings. Panel (b) shows the minimal energy spread as a function of the pickup coupling, both exactly from panel (a) and using the asymptotic formula σmf2≃32/34×21/3∣g1∣2+σm02∣g1∣4/3\sigma_{m_f}^2\simeq\frac{3^{2/3}}{4\times2^{1/3}}\frac{|g_1|^2+\sigma_{m_0}^2}{|g_1|^{4/3}}. The agreement is quite good despite the moderate values of both κ\kappa and ∣g1∣|g_1|. Cooling down to σmf≃1.425\sigma_{m_f}\simeq1.425 photons, less than 50% of the input spread, is possible. Panels (c) and (d) show the same plots for σm0=15\sigma_{m_0}=15. Cooling down to σmf≃2.3\sigma_{m_f}\simeq2.3 is possible in this case. Importantly, panels (b) and (d) show that while the optimum coupling is ∣g1∣≃2σm0|g_1|\simeq\sqrt{2}\sigma_{m_0}, significant cooling can still be achieved with much smaller coupling strengths. The optimal couplings are 4.24 and 21.2, respectively, while substantial cooling can still be accomplished with ∣g1∣=2|g_1|=2 and 5, respectively. Panel (e) reports the minimal final spread as a function of the initial spread alongside the asymptotic estimate σmf,opt≃35/38σm01/3\sigma_{m_f,\mathrm{opt}}\simeq\sqrt{\frac{3^{5/3}}{8}}\sigma_{m_0}^{1/3}, which agrees well. The dotted gray line shows the condition σmf=σm0\sigma_{m_f}=\sigma_{m_0} for reference. This shows that cooling is only possible if σm0≳1\sigma_{m_0}\gtrsim1, i.e., the initial energy spread exceeds the cooling photon energy.

Overview of single-pass cooling performance

Figure 2. Expected single-pass cooling behavior using analytic derivations. (a) and (c) show final energy spread as a function of pickup coupling and drift length for σm0=3\sigma_{m_0}=3 and 15, respectively. (b) and (d) show the optimal final energy spread as a function of the pickup coupling compared to asymptotic estimates. (e) shows the optimal final spread as a function of the initial spread alongside the asymptotic form σmf≃35/38σm01/3\sigma_{m_f}\simeq\sqrt{\frac{3^{5/3}}{8}}\sigma_{m_0}^{1/3}.

Numerical tracking

The analytic results of the last section demonstrate that significant energy spread reduction is possible, but it does not distinguish between true phase space cooling and conservative phase space manipulation. To that end, we numerically propagate the initial electron density matrix to validate the analytic results and evaluate the final phase space area. We represent the electron in the basis of momentum eigenstates with initial density matrix coefficients ρe,i(m,m′)\rho_{e,i}(m,m') and the photons in the vacuum state. We trace out the photon degrees of freedom from the final state as they will not be observed, yielding the final electron density matrix

ρe,f(m,m′)=∑jρe,i(m+j,m′+j)Sm+j,jSm′+j,j∗.(9)\rho_{e,f}(m,m')=\sum_j\rho_{e,i}(m+j,m'+j)S_{m+j,j}S^*_{m'+j,j}. \tag*{(9)}

where Sm,n=⟨n∣D^(g2)e−iκ(m−a^†a^)2D^(g1)∣0⟩S_{m,n}=\langle n|\hat{D}(g_2)e^{-i\kappa(m-\hat{a}^{\dagger}\hat{a})^2}\hat{D}(g_1)|0\rangle, with D^(g)=ega^†−g∗a^\hat{D}(g)=e^{g\hat{a}^{\dagger}-g^*\hat{a}} the normal bosonic displacement operator and ∣n⟩|n\rangle the nn excitation Fock state, see the SM for details. The final momentum spectrum is ρe,f(m,m)\rho_{e,f}(m,m), which depends only on the input energy spectrum and not on any off-diagonal coherence information, including any chirp in the initial phase space. Determining whether true cooling has been achieved can be done by evaluating the purity of the final state P=Tr⁡[ρ2]P=\operatorname{Tr}[\rho^2]. A pure state has P=1P = 1, while a highly mixed state has PP approaching zero. In this context, cooling means increasing the purity of the final state.

We consider an electron which starts in a statistical mixture of Gaussian states with coherent FWHM energy spread 0.1 eV, incoherent spread 0.7 eV, and incoherent FWHM duration 400 fs, similar to recent measurements from ultrafast electron microscopes [48]. Definitions and details of the implementation are provided in the SM. We consider a case with cooling photon energy ℏω=20\hbar\omega= 20 meV, corresponding to 4.84 THz light. The initial rms energy spread then corresponds to roughly 15 photons.

Figure 3 summarizes the numerical results. Panel (a) reports the optimal kicker coupling, pickup-kicker phase shift, and drift length as a function of ∣g1∣|g_1| per the analytic estimates. For strong pickup coupling, ∣g2∣|g_2| plateaus around 2, while κ\kappa gets smaller as more photons are emitted in the pickup. In panel (b), we evaluate the cooled energy spread and final state purity as a function of the pickup coupling using the numerical approach. Consistent with the analytic formulae, the optimal conditions are forgiving. Although the optimal ∣g1∣|g_1| is around 20, the final energy spread changes gradually for ∣g1∣≥5|g_1| \ge5. For ∣g1∣=5|g_1| = 5, the final FWHM energy spread is already around 150 meV while the purity has grown by more than a factor of four from the initial value of 6.4×10−36.4 \times10^{-3} up to 2.9×10−22.9 \times10^{-2}. This confirms that ∣g1∣opt≃2σm0|g_1|_{\mathrm{opt}} \simeq\sqrt{2}\sigma_{m0} is a conservative estimate, and significantly smaller couplings can be used.

Figure 3. Numerical evaluation of cooling a ∼0.7\sim0.7 eV FWHM spread with ℏω=20\hbar\omega= 20 meV photons. Panel (a) shows the optimal values of the kicker coupling, phase shift, and drift length as a function of pickup coupling. Panel (b) shows the final FWHM spread and state purity for different ∣g1∣|g_1|. Panels (c) and (d) show the initial and final Wigner distributions of the electron for ∣g1∣=5|g_1| = 5, with panel (e) showing the energy spectra.

Discussion — This single-pass approach relies on achieving coupling factors ∣g∣>1|g| > 1 at photon energies below the electron energy spread, generally sub-0.5 eV in modern microscopes. This implies operation in the mid-to-far-IR and THz regimes. Coupling between 0.1 and 1 has been achieved at optical frequencies [35, 42], and couplings above 1 are theoretically allowed but undemonstrated [30, 38, 40, 41]. Most work towards high coupling has focused in the optical and near-IR regimes because of overlap with available sources and interest in photon-induced near-field electron microscopy [49, 50, 51, 52], in which the photon energy should exceed the electron energy spread. On the other hand, designing and fabricating nanophotonic structures with high coupling in the far-IR and THz regimes should be easier, as the required feature sizes are larger and the near-fields decay over longer distances. For photon transport and optics between the pickup and kicker, this scheme may benefit from significant recent advances in THz [53, 54, 55] and CO2_2 laser [56, 57, 58] optics.

We have so far neglected optical losses, which may increase the required couplings and act as a source of decoherence. We consider lossy transport in the SM, finding corrected formulas for the optimal parameters as well as final energy spread σE≃0.88η−1/3(ℏω)2/3σE01/3\sigma_E \simeq0.88\eta^{-1/3}(\hbar\omega)^{2/3}\sigma_{E0}^{1/3} in the strong coupling regime, with η\eta the relative pickup-to-kicker power transmission. For the example in Figure 3, the final energy spread remains below 250 meV as long as the corrected values are used and the transmission stays above 30%. Lower couplings may be possible by introducing an amplifier between the pickup and kicker, which can help overcome losses and lower the required couplings. Although we focus here on single-pass cooling enabled by strong coupling, multi-pass operation with lower coupling could also be pursued in dedicated multi-pass electron microscopes [59, 60, 61], or in compact setups leveraging recent advances in on-chip electron resonators [62, 63]. Such multi-pass geometries could even explore transverse-longitudinal coupling optics to distribute the cooling across the whole six-dimensional phase space.

An important experimental consideration is the physical drift length required between the pickup and kicker. For the example explored above, κ≃0.01\kappa\simeq0.01 is required, corresponding to drift lengths of 1.4 cm, 21.5 cm, and 4.64 m at 5, 30, and 200 keV beam energies, respectively. This suggests that a free space drift could be sufficient for cooling in SEM geometries where the beam energy rarely exceeds 30 keV. For TEMs, temporarily lower energy sections or active dispersive elements, something analogous to the chicanes used in particle accelerators, could be considered.

In conclusion, we have shown that stochastic cooling, in which entropy is removed from the electron and left behind in a discarded optical mode, can be applied to electron microscopes. High-efficiency electron-light coupling in nanophotonic structures makes it possible to achieve significant cooling in a single-pass geometry compatible with electron microscope columns, and notably the scheme is laser-free. This requires vacuum coupling strengths ∣g∣≳1|g| \gtrsim1 in the mid-to-far IR or THz regimes along with reasonable (≳30%\gtrsim30\%) transport efficiency from the pickup to the kicker in a total length of a few tens of centimeters for SEM kinetic energies. In this case, cooling of the typical energy spread of typical modern sources around 0.7 eV FWHM down to the 100 meV level is achievable. While energy spread reduction is the most direct signature of the effect, it is reflected in a commensurate increase in the purity of the quantum state of the electron and a reduction of the longitudinal emittance. While we focused on using few-THz radiation to cool down to the 100 meV level, first experiments can leverage higher photon energies to demonstrate the concept. Multi-pass schemes can not only lower the required coupling coefficients, but also open the possibility of full six-dimensional phase space cooling. Overall, our work paves a path towards enhanced beam brightness in electron microscopes, even towards fundamental quantum limits.

Acknowledgments

This work was supported by the Gordon and Betty Moore Foundation (GBMF11473) and the DFG Leibniz Prize (DFG HO 4543/5-1). The authors thank Dr. Zhexin Zhao for helpful discussions and Hommelhoff group members for the support.

Supplemental Material

Derivation of the Scattering Operator

The general starting point for the interaction of the electron with a photonic mode is the minimally coupled Dirac Hamiltonian:

H=ℏωa^†a^+cα⋅(p^+eA^)+βmec2(10)H = \hbar\omega\hat{a}^{\dagger}\hat{a} + c\boldsymbol{\alpha}\cdot\left(\hat{\mathbf{p}} + e\hat{\mathbf{A}}\right) + \beta m_{\mathrm{e}}c^{2} \tag*{(10)}

where a^\hat{a} is the optical annihilation operator with ℏω\hbar\omega the photon energy, αi=(0σiσi0)\alpha_i = \begin{pmatrix} 0 & \sigma_i \\ \sigma_i & 0 \end{pmatrix} with σi\sigma_i the ith Pauli matrix, β=(I00−I)\beta= \begin{pmatrix} \mathbb{I} & 0 \\ 0 & -\mathbb{I} \end{pmatrix}, p^\hat{\mathbf{p}} the electron momentum operator, and A^\hat{\mathbf{A}} the quantized vector potential of the optical mode. mem_{\mathrm{e}} and cc are the electron mass and the speed of light, respectively. As shown in numerous other works [43, 44], if the electron energy spread is narrow around some central velocity v0v_0 in the zz direction, the optical field is weak so that ponderomotive effects can be neglected, and the electron does not diffract significantly over the interaction length, this can be reduced to the effective 1D Schrödinger Hamiltonian

H=v0p^+p^22γ03me+ℏωa^†a^+ev0A^z(z^)(11)H = v_0\hat{p} + \frac{\hat{p}^{2}}{2\gamma_0^{3}m_{\mathrm{e}}} + \hbar\omega\hat{a}^{\dagger}\hat{a} + ev_0\hat{A}_z(\hat{z}) \tag*{(11)}

where v0=c1−1γ02v_0 = c\sqrt{1-\frac{1}{\gamma_0^2}} and we now use p^\hat{p} to represent the zz momentum operator, specifically the deviation from the reference momentum γ0mev0\gamma_0m_{\mathrm{e}}v_0. The first two terms are the interaction-free energy of the electron, the third is the interaction-free energy of the optical mode, and the final term gives rise to interactions. We use the following convention for the optical vector potential:

A^z(z^)=iω(Ezeikz^a^−Ez∗e−ikz^a^†)(12)\hat{A}_z(\hat{z}) = \frac{i}{\omega}\left(E_z e^{ik\hat{z}}\hat{a} - E_z^{*}e^{-ik\hat{z}}\hat{a}^{\dagger}\right) \tag*{(12)}

where kk is the longitudinal wavenumber of the mode and EzE_z is the single-photon electric field. We move to the interaction picture with H0=v0p^+ℏωa^†a^H_0 = v_0\hat{p} + \hbar\omega\hat{a}^{\dagger}\hat{a}, yielding the interaction picture Hamiltonian

VI=p^22γ03me+iev0ω(Ezeikz^a^ei(kv0−ω)t−Ez∗e−ikz^a^†e−i(kv0−ω)t)(13)V_I = \frac{\hat{p}^{2}}{2\gamma_0^{3}m_{\mathrm{e}}} + \frac{iev_0}{\omega}\left(E_z e^{ik\hat{z}}\hat{a}e^{i(kv_0-\omega)t} - E_z^{*}e^{-ik\hat{z}}\hat{a}^{\dagger}e^{-i(kv_0-\omega)t}\right) \tag*{(13)}

For simplicity, we assume the electron-optical interaction is phase-matched with ω=kv0\omega= kv_0. In this case, the interaction picture Hamiltonian is time-independent, and the scattering operator is exactly

S^(g,κ)=exp⁡[−iκm^2+(ge−iκz^a^†−g∗eiκz^a^)](14)\hat{S}(g,\kappa) = \exp\left[-i\kappa\hat{m}^{2} + \left(ge^{-i\kappa\hat{z}}\hat{a}^{\dagger} - g^{*}e^{i\kappa\hat{z}}\hat{a}\right)\right] \tag*{(14)}

where we have introduced the scaled variables m^=p^/ℏk\hat{m}=\hat{p}/\hbar k, κ=Lℏk2/2γ03mev0\kappa=L\hbar k^{2}/2\gamma_{0}^{3}m_{e}v_{0}, and g=−eLEz∗/ℏωg=-eLE_{z}^{*}/\hbar\omega, with LL the interaction length. This scattering operator describes just as well a drift with no coupling by setting g=0g=0 as it does a coupling section. In most of the text we will take the no-recoil approximation in the pickup and kicker, in which κ\kappa is set to zero in the coupling regions. This leads to the scattering operator for the three-stage OSC transformation as we write it in the text:

S^T=S^(g2,0)S^(0,κ)S^(g1,0)(15)\hat{S}_{T}=\hat{S}(g_{2},0)\hat{S}(0,\kappa)\hat{S}(g_{1},0) \tag*{(15)}

AVERAGE PHOTON NUMBER AND VARIANCE

The scattering operator preserves the total excitation operator m^+a^†a^\hat{m}+\hat{a}^{\dagger}\hat{a}, reflecting energy-momentum conservation. It is most easily analyzed in the joint operator A^=a^eikz^\hat{A}=\hat{a}e^{ik\hat{z}} which simultaneously destroys a photon and increases the electron momentum by one quantum of photon momentum. Let ∣m0⟩⊗∣0⟩\lvert m_{0}\rangle\otimes\lvert0\rangle represent the separable state with electron momentum ℏkm0\hbar k m_{0} and the optical mode in the vacuum state. For this initial state, the scattering operator can only access states ∣m0−n⟩⊗∣n⟩\lvert m_{0}-n\rangle\otimes\lvert n\rangle with nn photons and electron momentum ℏk(m0−n)\hbar k(m_{0}-n). In this subspace, m^\hat{m} behaves identically to m0−A^†A^m_{0}-\hat{A}^{\dagger}\hat{A}. Note that A^†A^=a^†a^\hat{A}^{\dagger}\hat{A}=\hat{a}^{\dagger}\hat{a} simply counts the number of photons, and therefore the number of exchanged quanta, so the initial state behaves like the bosonic vacuum state. Within this subspace, we may write the scattering operator as

S^(g,κ)=exp⁡[−iκ(m0−A^†A^)2+(gA^†−g∗A^)](16)\hat{S}(g,\kappa)=\exp\left[-i\kappa\left(m_{0}-\hat{A}^{\dagger}\hat{A}\right)^{2}+\left(g\hat{A}^{\dagger}-g^{*}\hat{A}\right)\right] \tag*{(16)}

For κ=0\kappa=0, this is exactly the bosonic displacement operator D^(g)=egA^†−g∗A^\hat{D}(g)=e^{g\hat{A}^{\dagger}-g^{*}\hat{A}}. So, we may write the total scattering transformation within this subspace as

S^T(m0)=D^(g2)e−iκ(m0−A^†A^)2D^(g1)(17)\hat{S}_{T}(m_{0})=\hat{D}(g_{2})e^{-i\kappa\left(m_{0}-\hat{A}^{\dagger}\hat{A}\right)^{2}}\hat{D}(g_{1}) \tag*{(17)}

We proceed in the Heisenberg picture and determine how the annihilation operator transforms under this operation. For convenience we define the transformed annihilation operator Z^(m0)=S^T†(m0)A^S^T(m0)\hat{Z}(m_{0})=\hat{S}_{T}^{\dagger}(m_{0)\hat{A}\hat{S}_{T}(m_{0})}. The displacement operator D^(g)\hat{D}(g) sends the annihilation operator to A^+g\hat{A}+g, so

Z^(m0)=D^†(g1)eiκ(m0−A^†A^)2(A^+g2)e−iκ(m0−A^†A^)2D^(g1)(18)\hat{Z}(m_{0})=\hat{D}^{\dagger}(g_{1})e^{i\kappa\left(m_{0}-\hat{A}^{\dagger}\hat{A}\right)^{2}}(\hat{A}+g_{2})e^{-i\kappa\left(m_{0}-\hat{A}^{\dagger}\hat{A}\right)^{2}}\hat{D}(g_{1}) \tag*{(18)}

Now using the normal-ordering identity A^f(A^†A^)=f(A^†A^+1)A^\hat{A}f(\hat{A}^{\dagger}\hat{A})=f(\hat{A}^{\dagger}\hat{A}+1)\hat{A}, we have

Z^(m0)=D^†(g1)(eiκ(2m0−1)e−2iκA^†A^A^+g2)D^(g1)(19)\hat{Z}(m_{0})=\hat{D}^{\dagger}(g_{1})\left(e^{i\kappa(2m_{0}-1)}e^{-2i\kappa\hat{A}^{\dagger}\hat{A}}\hat{A}+g_{2}\right)\hat{D}(g_{1}) \tag*{(19)}
=eiκ(2m0−1)e−2iκ(A^†+g1∗)(A^+g1)(A^+g1)+g2(20)=e^{i\kappa(2m_{0}-1)}e^{-2i\kappa(\hat{A}^{\dagger}+g_{1}^{*})(\hat{A}+g_{1})}(\hat{A}+g_{1})+g_{2} \tag*{(20)}

Acting on the initial state, we have

Z^(m0)∣0⟩=(eiκ(2m0−1)e−2iκ(A^†+g1∗)(A^+g1)g1+g2)∣0⟩=(eiκ(2m0−1)D^†(g1)e−2iκA^†A^D^(g1)g1+g2)∣0⟩(21)\hat{Z}(m_{0})\lvert0\rangle=\left(e^{i\kappa(2m_{0}-1)}e^{-2i\kappa(\hat{A}^{\dagger}+g_{1}^{*})(\hat{A}+g_{1})}g_{1}+g_{2}\right)\lvert0\rangle=\left(e^{i\kappa(2m_{0}-1)}\hat{D}^{\dagger}(g_{1})e^{-2i\kappa\hat{A}^{\dagger}\hat{A}}\hat{D}(g_{1})g_{1}+g_{2}\right)\lvert0\rangle \tag*{(21)}

Then for example, the expectation value of the excitation number operator N^(m0)=Z^†(m0)Z^(m0)\hat{N}(m_{0})=\hat{Z}^{\dagger}(m_{0})\hat{Z}(m_{0}) is

⟨N^(m0)⟩=∣g1∣2+∣g2∣2+(g1g2∗eiκ(2m0−1)⟨g1∣e−2iκA^†A^∣g1⟩+c.c.)(22)\langle\hat{N}(m_{0})\rangle=|g_{1}|^{2}+|g_{2}|^{2}+\left(g_{1}g_{2}^{*}e^{i\kappa(2m_{0}-1)}\langle g_{1}|e^{-2i\kappa\hat{A}^{\dagger}\hat{A}}|g_{1}\rangle+\mathrm{c.c.}\right) \tag*{(22)}
=∣g1∣2+∣g2∣2+(g1g2∗eiκ(2m0−1)e−i∣g1∣2sin⁡(2κ)−2∣g1∣2sin⁡2(κ)+c.c.)(23)=|g_{1}|^{2}+|g_{2}|^{2}+\left(g_{1}g_{2}^{*}e^{i\kappa(2m_{0}-1)}e^{-i|g_{1}|^{2}\sin(2\kappa)}-2|g_{1}|^{2}\sin^{2}(\kappa)+\mathrm{c.c.}\right) \tag*{(23)}
=∣g1∣2+∣g2∣2+2∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)cos⁡(2κm0+ψ)(24)=|g_{1}|^{2}+|g_{2}|^{2}+2|g_{1}||g_{2}|e^{-2|g_{1}|^{2}\sin^{2}(\kappa)}\cos(2\kappa m_{0}+\psi) \tag*{(24)}

where we have defined ψ=Δϕg−κ−∣g1∣2sin⁡(2κ)\psi=\Delta\phi_{g}-\kappa-|g_{1}|^{2}\sin(2\kappa) with g1g2∗=∣g1∣∣g2∣eiΔϕgg_{1}g_{2}^{*}=|g_{1}||g_{2}|e^{i\Delta\phi_{g}}. Now, for the variance we can also evaluate Z^2∣0⟩\hat{Z}^{2}\lvert0\rangle:

Z^2(m0)∣0⟩=D^†(g1)(g22+eiκ(2m0−1)e−2iκA^†A^A^(g12eiκ(2m0−3)e−2iκA^†A^+2g1g2))D^(g1)∣0⟩(25)\hat{Z}^{2}(m_{0})\lvert0\rangle=\hat{D}^{\dagger}(g_{1})\left(g_{2}^{2}+e^{i\kappa(2m_{0}-1)}e^{-2i\kappa\hat{A}^{\dagger}\hat{A}}\hat{A}\left(g_{1}^{2}e^{i\kappa(2m_{0}-3)}e^{-2i\kappa\hat{A}^{\dagger}\hat{A}}+2g_{1}g_{2}\right)\right)\hat{D}(g_{1})\lvert0\rangle \tag*{(25)}

Defining the phase Φ(m0)≡2κm0+ψ\Phi(m_{0})\equiv2\kappa m_{0}+\psi, the expectation value of Z^†2Z^2\hat{Z}^{\dagger2}\hat{Z}^{2} follows as

⟨Z^†2Z^2⟩=∣g2∣4+4∣g1∣2∣g2∣2+∣g1∣4+4∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)[∣g2∣2cos⁡(Φ)+∣g1∣2cos⁡(Φ−2κ)]+2∣g1∣2∣g2∣2e−2∣g1∣2sin⁡2(2κ)cos⁡(2Φ−2κ+8∣g1∣2cos⁡(κ)sin⁡3(κ))(26)\langle\hat{Z}^{\dagger2}\hat{Z}^{2}\rangle=|g_{2}|^{4}+4|g_{1}|^{2}|g_{2}|^{2}+|g_{1}|^{4}+4|g_{1}||g_{2}|e^{-2|g_{1}|^{2}\sin^{2}(\kappa)}\left[|g_{2}|^{2}\cos(\Phi)+|g_{1}|^{2}\cos(\Phi-2\kappa)\right] +2|g_{1}|^{2}|g_{2}|^{2}e^{-2|g_{1}|^{2}\sin^{2}(2\kappa)}\cos\left(2\Phi-2\kappa+8|g_{1}|^{2}\cos(\kappa)\sin^{3}(\kappa)\right) \tag*{(26)}

The photon number fluctuation is

σneff2(m0)=⟨Z^†2Z^2⟩+⟨N^⟩−⟨N^⟩2(27)\sigma_{n_{\mathrm{eff}}}^{2}(m_{0})=\langle\hat{Z}^{\dagger2}\hat{Z}^{2}\rangle+\langle\hat{N}\rangle-\langle\hat{N}\rangle^{2} \tag*{(27)}
=⟨N^(m0)⟩+2∣g1∣2∣g2∣2[1−2e−4∣g1∣2sin⁡2(κ)cos⁡2(Φ)+e−2∣g1∣2sin⁡2(2κ)cos⁡(2Φ−2κ+8∣g1∣2cos⁡(κ)sin⁡3(κ))]+4∣g1∣3∣g2∣e−2∣g1∣2sin⁡2(κ)[cos⁡(Φ−2κ)−cos⁡(Φ)](28)=\langle\hat{N}(m_{0})\rangle+2|g_{1}|^{2}|g_{2}|^{2}\left[1-2e^{-4|g_{1}|^{2}\sin^{2}(\kappa)}\cos^{2}(\Phi)+e^{-2|g_{1}|^{2}\sin^{2}(2\kappa)}\cos\left(2\Phi-2\kappa+8|g_{1}|^{2}\cos(\kappa)\sin^{3}(\kappa)\right)\right] +4|g_{1}|^{3}|g_{2}|e^{-2|g_{1}|^{2}\sin^{2}(\kappa)}\left[\cos(\Phi-2\kappa)-\cos(\Phi)\right] \tag*{(28)}

The first term is simply Poissonian noise; the final term arises from recoil in the drift and vanishes for κ→0\kappa\to0.

The results so far are valid for an electron sitting initially in a momentum eigenstate. The final scaled momentum of the electron is given by mf=m0−neff(m0)m_f=m_0-n_{\mathrm{eff}}(m_0), and we would now like to understand how the initial statistical distribution of m0m_0 translates into the final distribution of mfm_f. We have

Var⁡(mf)=Var⁡(m0)+Var⁡(neff)−2Cov⁡(m0,neff(m0))(29)\operatorname{Var}(m_f)=\operatorname{Var}(m_0)+\operatorname{Var}(n_{\mathrm{eff}})-2\operatorname{Cov}(m_0,n_{\mathrm{eff}}(m_0)) \tag*{(29)}

Now, the variance of neffn_{\mathrm{eff}} follows from the law of total variance:

Var⁡(neff)=E[Var⁡(neff∣m0)]+Var⁡(E[neff∣m0])(30)\operatorname{Var}(n_{\mathrm{eff}})=E[\operatorname{Var}(n_{\mathrm{eff}}|m_0)]+\operatorname{Var}(E[n_{\mathrm{eff}}|m_0]) \tag*{(30)}

In a fully classical calculation, Var⁡(neff∣m0)\operatorname{Var}(n_{\mathrm{eff}}|m_0) is zero, so this first term accounts for quantum fluctuations in the photon emission process. The second term, on the other hand, accounts for the electron emitting different average photon numbers depending on its initial energy and should recover classical results. E[neff∣m0]E[n_{\mathrm{eff}}|m_0] is exactly ⟨neff(m0)⟩\langle n_{\mathrm{eff}}(m_0)\rangle that we calculated before, while Var⁡(neff∣m0)\operatorname{Var}(n_{\mathrm{eff}}|m_0) is σneff2(m0)\sigma_{n_{\mathrm{eff}}}^{2}(m_0). The remaining expectation value and variance must be taken over the statistical distribution of m0m_0, which we assume to be Gaussian according to f(m0)=12πσm02e−m022σm02f(m_0)=\frac{1}{\sqrt{2\pi\sigma_{m_0}^{2}}}e^{-\frac{m_0^{2}}{2\sigma_{m_0}^{2}}}. This yields, for the classical part,

Var⁡(E[neff∣m0])=2∣g1∣2∣g2∣2e−4∣g1∣2sin⁡2(κ)(1−e−4κ2σm02)(1−e−4κ2σm02cos⁡(2ψ))(31)\operatorname{Var}(E[n_{\mathrm{eff}}|m_0])=2|g_1|^{2}|g_2|^{2}e^{-4|g_1|^{2}\sin^{2}(\kappa)}\left(1-e^{-4\kappa^{2}\sigma_{m_0}^{2}}\right)\left(1-e^{-4\kappa^{2}\sigma_{m_0}^{2}}\cos(2\psi)\right) \tag*{(31)}

and

E[Var⁡(neff∣m0)]=∣g1∣2+∣g2∣2+2∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)−2κ2σm02cos⁡(ψ)+2∣g1∣2∣g2∣2[1−e−4∣g1∣2sin⁡2(κ)(1+e−8κ2σm02cos⁡(2ψ))]+e−2∣g1∣2sin⁡2(2κ)e−8κ2σm02cos⁡(2ψ−2κ+8∣g1∣2cos⁡(κ)sin⁡3(κ))+4∣g1∣3∣g2∣e−2∣g1∣2sin⁡2(κ)e−2κ2σm02[cos⁡(ψ−2κ)−cos⁡(ψ)](32)\begin{aligned} E[\operatorname{Var}(n_{\mathrm{eff}}|m_0)]={}&|g_1|^{2}+|g_2|^{2}+2|g_1||g_2|e^{-2|g_1|^{2}\sin^{2}(\kappa)-2\kappa^{2}\sigma_{m_0}^{2}}\cos(\psi)\\ &+2|g_1|^{2}|g_2|^{2}\left[1-e^{-4|g_1|^{2}\sin^{2}(\kappa)}\left(1+e^{-8\kappa^{2}\sigma_{m_0}^{2}}\cos(2\psi)\right)\right]\\ &+e^{-2|g_1|^{2}\sin^{2}(2\kappa)}e^{-8\kappa^{2}\sigma_{m_0}^{2}}\cos\left(2\psi-2\kappa+8|g_1|^{2}\cos(\kappa)\sin^{3}(\kappa)\right)\\ &+4|g_1|^{3}|g_2|e^{-2|g_1|^{2}\sin^{2}(\kappa)}e^{-2\kappa^{2}\sigma_{m_0}^{2}}\left[\cos(\psi-2\kappa)-\cos(\psi)\right] \tag*{(32)} \end{aligned}

The covariance Cov⁡(m0,neff(m0))\operatorname{Cov}(m_0,n_{\mathrm{eff}}(m_0)) can be calculated from Stein’s Lemma [64], which says that for a normally distributed random variable, Cov⁡(m0,neff(m0))=Var⁡(m0)E[neff′(m0)]\operatorname{Cov}(m_0,n_{\mathrm{eff}}(m_0))=\operatorname{Var}(m_0)E[n'_{\mathrm{eff}}(m_0)], where

E[neff′(m0)]=−4κ∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)e−2κ2σm02sin⁡(ψ)(33)E[n'_{\mathrm{eff}}(m_0)]=-4\kappa|g_1||g_2|e^{-2|g_1|^{2}\sin^{2}(\kappa)}e^{-2\kappa^{2}\sigma_{m_0}^{2}}\sin(\psi) \tag*{(33)}

A particularly instructive way to write the final energy spread is

σmf2=σm02(1+4κ∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)−2κ2σm02sin⁡(ψ))2+2∣g1∣2∣g2∣2e−4∣g1∣2sin⁡2(κ)[(1−e−4κ2σm02)(1−e−4κ2σm02cos⁡(2ψ))−8σm02κ2e−4κ2σm02sin⁡2(ψ)]+E[Var⁡(neff∣m0)](34)\begin{aligned} \sigma_{m_f}^{2}={}&\sigma_{m_0}^{2}\left(1+4\kappa|g_1||g_2|e^{-2|g_1|^{2}\sin^{2}(\kappa)-2\kappa^{2}\sigma_{m_0}^{2}}\sin(\psi)\right)^{2}\\ &+2|g_1|^{2}|g_2|^{2}e^{-4|g_1|^{2}\sin^{2}(\kappa)}\left[\left(1-e^{-4\kappa^{2}\sigma_{m_0}^{2}}\right)\left(1-e^{-4\kappa^{2}\sigma_{m_0}^{2}}\cos(2\psi)\right)-8\sigma_{m_0}^{2}\kappa^{2}e^{-4\kappa^{2}\sigma_{m_0}^{2}}\sin^{2}(\psi)\right]\\ &+E[\operatorname{Var}(n_{\mathrm{eff}}|m_0)] \tag*{(34)} \end{aligned}

With this decomposition, the first term motivates the linear-order cooling condition 4κ∣g1∣∣g2∣e−2∣g1∣2sin⁡2(κ)−2κ2σm02sin⁡(ψ)=−14\kappa|g_1||g_2|e^{-2|g_1|^{2}\sin^{2}(\kappa)-2\kappa^{2}\sigma_{m_0}^{2}}\sin(\psi)=-1 as discussed in the text, with a correction e−2κ2σm02e^{-2\kappa^{2}\sigma_{m_0}^{2}}. The second term reflects the residual phase space nonlinearity after cancellation of the first term, and the third term reflects the quantum mechanical fluctuations of the emitted photon number. This exact expression is quadratic in the kicker coupling ∣g2∣|g_2|, so there is an explicit optimal value in terms of arbitrary ∣g1∣|g_1|, κ\kappa, and ψ\psi. The complex dependence on the remaining parameters prevents further analytical optimization without some assumptions.

Equation (34) is exact within the approximations leading to our scattering operator. We can use it to numerically determine optimal cooling conditions in the place of full numerical density matrix tracking. Before that, we can introduce some simplifications and approximations to suggest useful working points. Just assuming ψ=−π/2\psi=-\pi/2 yields the still exact but greatly simplified expression

σmf2=σm02+∣g1∣2−8κ∣g1∣∣g2∣e−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)]+∣g2∣2[1+2∣g1∣2(1−e−8κ2σmeff2+8∣g1∣2sin⁡4(κ)cos⁡(2κ−8∣g1∣2cos⁡(κ)sin⁡3(κ)))](35)\sigma_{m_f}^{2}=\sigma_{m_0}^{2}+|g_1|^{2}-8\kappa|g_1||g_2|e^{-2\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}\left[\sigma_{m_0}^{2}+|g_1|^{2}\operatorname{sinc}(\kappa)\cos(\kappa)\right]+|g_2|^{2}\left[1+2|g_1|^{2}\left(1-e^{-8\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}+8|g_1|^{2}\sin^{4}(\kappa)}\cos\left(2\kappa-8|g_1|^{2}\cos(\kappa)\sin^{3}(\kappa)\right)\right)\right] \tag*{(35)}

where σmeff=σm02+∣g1∣2sinc⁡2(κ)\sigma_{m_{\mathrm{eff}}}=\sqrt{\sigma_{m_0}^{2}+|g_1|^{2}\operatorname{sinc}^{2}(\kappa)}. In this form, it becomes clear that there is an unambiguous optimal kicker coupling

∣g2∣opt=4κ∣g1∣e−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)]1+2∣g1∣2(1−e−8κ2σmeff2+8∣g1∣2sin⁡4(κ)cos⁡(2κ−8∣g1∣2cos⁡(κ)sin⁡3(κ)))(36)|g_2|_{\mathrm{opt}}=\frac{4\kappa|g_1|e^{-2\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}\left[\sigma_{m_0}^{2}+|g_1|^{2}\operatorname{sinc}(\kappa)\cos(\kappa)\right]}{1+2|g_1|^{2}\left(1-e^{-8\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}+8|g_1|^{2}\sin^{4}(\kappa)}\cos\left(2\kappa-8|g_1|^{2}\cos(\kappa)\sin^{3}(\kappa)\right)\right)} \tag*{(36)}

Evaluated at this optimal value,

σmf2(∣g2∣opt)=σm02+∣g1∣2−4κ∣g1∣∣g2∣e−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)](37)\sigma_{m_f}^{2}(|g_2|_{\mathrm{opt}})=\sigma_{m_0}^{2}+|g_1|^{2}-4\kappa|g_1||g_2|e^{-2\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}\left[\sigma_{m_0}^{2}+|g_1|^{2}\operatorname{sinc}(\kappa)\cos(\kappa)\right] \tag*{(37)}

In the high-coupling regime, κ\kappa generally remains much less than 1, in which case we can approximate

∣g2∣opt≃4κσmeff2∣g1∣e−2κ2σmeff21+2∣g1∣2(1−e−8κ2σmeff2)(38)|g_2|_{\mathrm{opt}}\simeq\frac{4\kappa\sigma_{m_{\mathrm{eff}}}^{2}|g_1|e^{-2\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}}{1+2|g_1|^{2}\left(1-e^{-8\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}\right)} \tag*{(38)}

and

σmf2(∣g2∣opt)≃σmeff2[1−16κ2σmeff2∣g1∣2cosh⁡(4κ2σmeff2)+(1+4∣g1∣2)sinh⁡(4κ2σmeff2)](39)\sigma_{m_f}^{2}(|g_2|_{\mathrm{opt}})\simeq\sigma_{m_{\mathrm{eff}}}^{2}\left[1-\frac{16\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}|g_1|^{2}}{\cosh(4\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2})+(1+4|g_1|^{2})\sinh(4\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2})}\right] \tag*{(39)}

Defining x=4κ2σmeff2x=4\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2} and G=1+4∣g1∣2G=1+4|g_1|^{2}, this is minimized as a function of xx when

xG−1G−x=tanh⁡(x)(40)\frac{xG-1}{G-x}=\tanh(x) \tag*{(40)}

Now, for κ≪1\kappa\ll1 we can estimate tanh⁡(x)≃x−x33\tanh(x)\simeq x-\frac{x^{3}}{3}, while

Gx−1G−x≃−1G+(1−1G2)x+(1G−1G3)x2≃−1G+x(41)\frac{Gx-1}{G-x}\simeq-\frac{1}{G}+\left(1-\frac{1}{G^{2}}\right)x+\left(\frac{1}{G}-\frac{1}{G^{3}}\right)x^{2}\simeq-\frac{1}{G}+x \tag*{(41)}

where we’ve assumed both that x≪1x\ll1 and G≫1G\gg1. The condition then reduces to x∗=(3/G)1/3x^{*}=(3/G)^{1/3}. This predicts the optimal drift

κopt≃12σmeff(31+4∣g1∣2)1/6(42)\kappa_{\mathrm{opt}}\simeq\frac{1}{2\sigma_{m_{\mathrm{eff}}}}\left(\frac{3}{1+4|g_1|^{2}}\right)^{1/6} \tag*{(42)}

Plugging this into the energy spread and taking ∣g1∣≫1|g_1|\gg1 leads to the lowest-order expression,

σmf2(∣g2∣opt,κopt)≃32/34×21/3σm02+∣g1∣2∣g1∣4/3(43)\sigma_{m_f}^{2}(|g_2|_{\mathrm{opt}},\kappa_{\mathrm{opt}})\simeq\frac{3^{2/3}}{4\times2^{1/3}}\frac{\sigma_{m_0}^{2}+|g_1|^{2}}{|g_1|^{4/3}} \tag*{(43)}

This is minimized as a function of ∣g1∣|g_1| for ∣g1∣opt=2σm0|g_1|_{\mathrm{opt}}=\sqrt{2}\sigma_{m_0}, for which it takes the value σmf,opt=35/38σm01/3\sigma_{m_f,\mathrm{opt}}=\sqrt{\frac{3^{5/3}}{8}}\sigma_{m_0}^{1/3}.

Description of mixed electron states

We consider the initial electronic state to be a statistical mixture of generally chirped Gaussian wavepackets. A chirped Gaussian wavepacket with central momentum ℏkm0\hbar k m_0, chirp CC, and scaled longitudinal position ϕ0=kz0\phi_0=k z_0 can be described by the superposition of momentum eigenstates

∣m0,ϕ0⟩=∫dm[12πσmc2]1/4exp⁡[−(m−m0)24σmc2(1+iC)−i(m−m0)ϕ0]∣m⟩≡∫dmψp(m;m0,ϕ0)∣m⟩(44)|m_0,\phi_0\rangle=\int dm\left[\frac{1}{2\pi\sigma_{m_c}^{2}}\right]^{1/4}\exp\left[-\frac{(m-m_0)^{2}}{4\sigma_{m_c}^{2}}(1+iC)-i(m-m_0)\phi_0\right]|m\rangle\equiv\int dm\psi_p(m;m_0,\phi_0)|m\rangle \tag*{(44)}

where σmc\sigma_{mc} is the coherent rms energy spread. Furthermore, we can describe the statistical fluctuations of the wavepackets using the distribution function

fs(m0,ϕ0)=12πϵmiexp⁡[−σϕi2m02−2σmiϕim0ϕ0+σmi2ϕ022ϵmi2](45)f_s(m_0,\phi_0)=\frac{1}{2\pi\epsilon_{mi}}\exp\left[-\frac{\sigma_{\phi_i}^{2}m_0^{2}-2\sigma_{mi\phi_i}m_0\phi_0+\sigma_{mi}^{2}\phi_0^{2}}{2\epsilon_{mi}^{2}}\right] \tag*{(45)}

where σmi\sigma_{mi} and σϕi\sigma_{\phi_i} are the incoherent energy and phase spreads, respectively, σmiϕi\sigma_{mi\phi_i} representing the incoherent chirp, and ϵmi2=σmi2σϕi2−σmiϕi2\epsilon_{mi}^{2}=\sigma_{mi}^{2}\sigma_{\phi_i}^{2}-\sigma_{mi\phi_i}^{2} the longitudinal emittance. The density matrix of the resulting statistical mixture of chirped wavepackets is

ρe=∫dm0dϕ0fs(m0,ϕ0)∣m0,ϕ0⟩⟨m0,ϕ0∣(46)\rho_e=\int dm_0d\phi_0 f_s(m_0,\phi_0)|m_0,\phi_0\rangle\langle m_0,\phi_0| \tag*{(46)}
=∫dm0dϕ0fs(m0,ϕ0)∫dmdm′ψp(m;m0,ϕ0)ψp∗(m′;m0,ϕ0)∣m⟩⟨m′∣(47)=\int dm_0d\phi_0 f_s(m_0,\phi_0)\int dmdm'\psi_p(m;m_0,\phi_0)\psi_p^*(m';m_0,\phi_0)|m\rangle\langle m'| \tag*{(47)}
=∫dmdm′ρe(m,m′)∣m⟩⟨m′∣(48)=\int dmdm'\rho_e(m,m')|m\rangle\langle m'| \tag*{(48)}

where the density matrix elements are

ρe(m,m′)=12π(σmc2+σmi2)exp⁡[−(m+m′)2+2i(m2−m′2)(C+2σmiϕi)8(σmc2+σmi2)]\rho_e(m,m')=\frac{1}{\sqrt{2\pi(\sigma_{mc}^{2}+\sigma_{mi}^{2})}}\exp\left[-\frac{(m+m')^{2}+2i(m^{2}-m'^{2})(C+2\sigma_{mi\phi_i})}{8(\sigma_{mc}^{2}+\sigma_{mi}^{2})}\right]
×exp⁡[−(m−m′)2((1+C2)σmc2+4σmc4σϕi2+σmc2(1+4ϵmi2−4Cσmiϕi))8σmc2(σmc2+σmi2)](49){}\times\exp\left[-\frac{(m-m')^{2}\left((1+C^{2})\sigma_{mc}^{2}+4\sigma_{mc}^{4}\sigma_{\phi_i}^{2}+\sigma_{mc}^{2}(1+4\epsilon_{mi}^{2}-4C\sigma_{mi\phi_i})\right)}{8\sigma_{mc}^{2}(\sigma_{mc}^{2}+\sigma_{mi}^{2})}\right] \tag*{(49)}

Though we include the chirp parameter here, it does not impact the spectrum ρ(m,m)\rho(m,m) and therefore plays no role in the single-pass cooling of the energy spread.

Multi-electron effects and valid current range

Our analysis is valid when a single electron is active in the cooling system at a time. Multi-electron effects become important when the pickup radiation emitted by one electron can interact with another in the kicker. In that case, the radiation from one electron acts as an incoherent kick on the second. Avoiding this requires that a single electron’s radiation pulse duration is shorter than the inter-electron spacing. In a periodic nanostructure, the radiated pulse length is approximately the number of periods multiplied by the radiation wavelength. Taking for example Smith-Purcell-like radiators, the phase-matching condition for radiation emitted normal to the electron propagation direction requires β0=Λ/λ\beta_0=\Lambda/\lambda, with λ\lambda the radiation wavelength and Λ\Lambda the structure period. The radiation pulse duration is then (L/Λ)λ/c=L/v0(L/\Lambda)\lambda/c=L/v_0. The average separation between electrons in a beam with current II is e/Ie/I. The single-active-electron condition is then IL≪ev0IL\ll ev_0. This implies an upper limit of IL≃1.58IL\simeq1.58 nA⋅\cdotcm at 30 keV kinetic energy, and IL≃3.34IL\simeq3.34 nA⋅\cdotcm at 200 keV. Assuming structure lengths less than 1 cm, currents in the nA range should pose no issue, and higher currents can be possible depending on the exact design. Cooling is still possible with multi-electron effects included, just at a reduced rate. A detailed analysis of these effects is beyond our current scope.

Recoil effects in pickup and kicker and their compensation

In the analysis thus far we have ignored the effects of recoil in the pickup and kicker. With such large couplings, it could be important to take this into account. We can do so by using the full scattering matrix:

S^T=S^(g2,κ2)S^(0,κ)S^(g1,κ1)(50)\hat{S}_T=\hat{S}(g_2,\kappa_2)\hat{S}(0,\kappa)\hat{S}(g_1,\kappa_1) \tag*{(50)}
=e−iκ2m^2+(g2b^a^†−g2b^†a^)e−iκm^2e−iκ1m^2+(g1b^a^†−g1b^†a^)(51)=e^{-i\kappa_2\hat{m}^{2}+(g_2\hat{b}\hat{a}^{\dagger}-g_2\hat{b}^{\dagger}\hat{a})}e^{-i\kappa\hat{m}^{2}}e^{-i\kappa_1\hat{m}^{2}+(g_1\hat{b}\hat{a}^{\dagger}-g_1\hat{b}^{\dagger}\hat{a})} \tag*{(51)}

The inclusion of recoil in the pickup and kicker has two effects. The lowest-order effect is simply to lengthen the effective drift from κ\kappa to something like κ+κ1+κ2\kappa+\kappa_1+\kappa_2. This is in fact an overestimate, since the electron accumulates energy changes over the length of the structure. A reasonable estimate for the effective increased drift length might be κeff=κ+(κ1+κ2)/2\kappa_{\mathrm{eff}}=\kappa+(\kappa_1+\kappa_2)/2. Beyond an effective drift length correction, recoil can also qualitatively change the behavior of the electron-light interaction in the pickup and kicker [65, 66, 67, 68, 69]. The first effect can be compensated simply by shifting the κ\kappa used in the drift by the correction term (κ1+κ2)/2(\kappa_1+\kappa_2)/2.

Figure 4 showcases the effects of including recoil in the numerical calculations and their compensation with a corrected drift length. The parameters are the same as in the main text with ∣g1∣=5|g_1|=5, only now we consider identical structure lengths for simplicity with κ1=κ2\kappa_1=\kappa_2, which can be converted to physical lengths when a kinetic energy is specified. Panels (a) and (b) show the final energy spectrum as a function of the structure length. In (a), the drift length is uncorrected and for longer structure lengths, the spectrum shifts further from the nominal position and becomes distorted. In panel (b), we reduce the drift length by (κ1+κ2)/2(\kappa_1+\kappa_2)/2 to compensate for the extra dispersion in the pickup and kicker. We see that the spectrum is much less affected by the pickup and kicker recoil in this case. Panel (c) shows the final FWHM energy spread and state purity as a function of the structure lengths. We also plot the true structure length at 30 keV kinetic energy on the top x axis for reference. In the uncorrected case, the final spread grows and the purity shrinks for L>2L>2 cm. When the drift length is corrected, this degradation is significantly weaker, and the tolerable structure length can be extended to ≃40\simeq40 cm. Beyond this, the corrected drift length drops below zero, becoming unphysical. 40 cm structure length is already impractical anyways, so we conclude that for any practical design, recoil effects are easily correctable with small adjustments to the drift length.

Figure 4. Recoil effects in the pickup and kicker, and their compensation with effective drift correction for κ1=κ2\kappa_1=\kappa_2. Panels (a) and (b) show the final spectrum as a function of the scaled structure length without and with drift length compensation. Panel (c) shows the final FWHM energy spread and state purity as a function of the same.

For further reference, in Figure 5 we independently scan κ1\kappa_1 and κ2\kappa_2 and plot the resulting purity of the state with and without recoil correction. The white dashed line marks the region beyond which the corrected drift becomes negative. Again we see that the purity is minimally affected when correction is applied for a large range of pickup and kicker scaled lengths. The performance is more sensitive to recoil in the pickup than the kicker, but across the entire region the purity is maintained above at least 75% of the no-recoil value. We also note that our correction (κ1+κ2)/2(\kappa_1+\kappa_2)/2 is a simple estimate, and even better performance could be possible with careful optimization.

Figure 5. Final purity as a function of scaled lengths of the pickup (κ1\kappa_1) and kicker (κ2\kappa_2) with and without recoil correction of the drift length.

Impact of photon loss during transport

Loss of photons during transport from the pickup to the kicker can be modeled generally with the beam-splitter transformation (see, for example, [70, 71])

a^→ηa^+1−ηF^(52)\hat{a} \to\sqrt{\eta}\hat{a}+\sqrt{1-\eta}\hat{F} \tag*{(52)}

where F^\hat{F} is the annihilation operator for a fictitious bosonic loss channel starting in the vacuum state, with unitarity requiring [F^,F^†]=1[\hat{F},\hat{F}^{\dagger}]=1. This is represented by the unitary transformation

S^L=eθ(a^†F^−a^F^†)(53)\hat{S}_{L}=e^{\theta(\hat{a}^{\dagger}\hat{F}-\hat{a}\hat{F}^{\dagger})} \tag*{(53)}

where cos⁡(θ)=η\cos(\theta)=\sqrt{\eta}. The total transformation is then S^T=S^(g2,0)S^LS^(0,κ)S^(g1,0)\hat{S}_{T}=\hat{S}(g_{2},0)\hat{S}_{L}\hat{S}(0,\kappa)\hat{S}(g_{1},0), and since S^L\hat{S}_{L} commutes with the drift transformation we can think of it as being applied either after the pickup or before the kicker. If we take the latter, we can note that

S^(g2,0)S^L=S^LS^L†S^(g2,0)S^L=S^Lexp⁡[g2b^†(ηa^†+1−ηF^†)−H.c.](54)\hat{S}(g_{2},0)\hat{S}_{L}=\hat{S}_{L}\hat{S}_{L}^{\dagger}\hat{S}(g_{2},0)\hat{S}_{L}=\hat{S}_{L}\exp\left[g_{2}\hat{b}^{\dagger}\left(\sqrt{\eta}\hat{a}^{\dagger}+\sqrt{1-\eta}\hat{F}^{\dagger}\right)-\mathrm{H.c.}\right] \tag*{(54)}

Now, b^\hat{b}, a^\hat{a}, and F^\hat{F} all commute with each other, and further b^\hat{b} commutes with b^†\hat{b}^{\dagger}, so this can be factorized as

S^(g2,0)S^L=S^LS^F(1−ηg2,0)S^(ηg2,0)(55)\hat{S}(g_{2},0)\hat{S}_{L}=\hat{S}_{L}\hat{S}_{F}\left(\sqrt{1-\eta}g_{2},0\right)\hat{S}\left(\sqrt{\eta}g_{2},0\right) \tag*{(55)}

where we’ve written S^F(g,0)\hat{S}_{F}(g,0) as equivalent to S^(g,0)\hat{S}(g,0), only acting on the F^\hat{F} subspace instead of a^\hat{a}.

In this form, the total transformation before application of S^F\hat{S}_{F} is exactly the same as the original one, only now with g2g_{2} scaled to an effective value ηg2\sqrt{\eta}g_{2}. The whole transformation is then followed by another unitary transformation which mixes the electron-photon Hilbert space with the Hilbert space of the lossy mode, leading to decoherence. We may write the total lossy scattering operator as

S^T,L=S^LS^F(1−ηg2,0)S^T(g1,ηg2,κ)(56)\hat{S}_{T,L}=\hat{S}_{L}\hat{S}_{F}\left(\sqrt{1-\eta}g_{2},0\right)\hat{S}_{T}\left(g_{1},\sqrt{\eta}g_{2},\kappa\right) \tag*{(56)}

The density matrix at the end of the loss-corrected OSC transformation is an entangled state between the electrons and the primary photonic mode, separably multiplied by the vacuum state for the loss mode ∣0⟩⟨0∣F|0\rangle\langle0|_{F}. The final electronic state is the trace over both a^\hat{a} and F^\hat{F} degrees of freedom:

ρe,f=Tr⁡a,F[S^LS^F(ρe,a⊗∣0⟩⟨0∣F)S^F†S^L†]=Tr⁡a,F[S^F(ρe,a⊗∣0⟩⟨0∣F)S^F†](57)\rho_{e,f}=\operatorname{Tr}_{a,F}\left[\hat{S}_{L}\hat{S}_{F}\left(\rho_{e,a}\otimes|0\rangle\langle0|_{F}\right)\hat{S}_{F}^{\dagger}\hat{S}_{L}^{\dagger}\right]=\operatorname{Tr}_{a,F}\left[\hat{S}_{F}\left(\rho_{e,a}\otimes|0\rangle\langle0|_{F}\right)\hat{S}_{F}^{\dagger}\right] \tag*{(57)}

S^F\hat{S}_{F} is completely independent of a^\hat{a}, so the trace over a^\hat{a} passes directly onto the density matrix, yielding the photon-traced-out electronic density matrix of the normal OSC transformation with scaled ∣g2∣|g_{2}|. Writing this as ρe,f=∫dmdm′ρe,f(m,m′)∣m⟩⟨m′∣\rho_{e,f}=\int dmdm'\rho_{e,f}(m,m')|m\rangle\langle m'|, and letting S^F∣m⟩⊗∣0⟩F=∑lDl,0∣m−l⟩⊗∣l⟩F\hat{S}_{F}|m\rangle\otimes|0\rangle_{F}=\sum_{l}D_{l,0}|m-l\rangle\otimes|l\rangle_{F}, we find the final electronic density matrix after tracing out the loss subsystem

ρe,f=∫dmdm′[∑lρ(m+l,m′+l)∣Dl,0(g21−η)∣2]∣m⟩⟨m′∣(58)\rho_{e,f}=\int dmdm'\left[\sum_{l}\rho(m+l,m'+l)\left|D_{l,0}\left(g_{2}\sqrt{1-\eta}\right)\right|^{2}\right]|m\rangle\langle m'| \tag*{(58)}

Conveniently, the matrix elements for the displacement operator starting from vacuum have a simple closed form:

∣Dl,0(α)∣2=e−∣α∣2∣α∣2ll!(59)|D_{l,0}(\alpha)|^{2}=e^{-|\alpha|^{2}}\frac{|\alpha|^{2l}}{l!} \tag*{(59)}

so we have

ρe,f=∫dmdm′[∑lρ(m+l,m′+l)e−∣g2∣2(1−η)(∣g2∣2(1−η))ll!]∣m⟩⟨m′∣(60)\rho_{e,f}=\int dmdm'\left[\sum_{l}\rho(m+l,m'+l)e^{-|g_{2}|^{2}(1-\eta)}\frac{\left(|g_{2}|^{2}(1-\eta)\right)^{l}}{l!}\right]|m\rangle\langle m'| \tag*{(60)}

The final result is a discrete convolution of the final OSC density matrix with a Poissonian of mean (1−η)∣g2∣2(1-\eta)|g_{2}|^{2}. Such a convolution simply shifts the mean of the spectrum by (1−η)∣g2∣2(1-\eta)|g_{2}|^{2} and adds the same to the final variance. It follows that the final energy spread is, for ψ=−π/2\psi=-\pi/2,

σmf2=σm02+∣g1∣2−8κ∣g1∣η∣g2∣e−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)]+η∣g2∣2[1η+2∣g1∣2(1−e−8κ2σmeff2+8∣g1∣2sin⁡4(κ)cos⁡(2κ−8∣g1∣2cos⁡(κ)sin⁡3(κ)))](61)\sigma_{m_f}^{2}=\sigma_{m_0}^{2}+|g_{1}|^{2}-8\kappa|g_{1}|\sqrt{\eta}|g_{2}|e^{-2\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}}\left[\sigma_{m_0}^{2}+|g_{1}|^{2}\operatorname{sinc}(\kappa)\cos(\kappa)\right] +\eta|g_{2}|^{2}\left[\frac{1}{\eta}+2|g_{1}|^{2}\left(1-e^{-8\kappa^{2}\sigma_{m_{\mathrm{eff}}}^{2}+8|g_{1}|^{2}\sin^{4}(\kappa)}\cos(2\kappa-8|g_{1}|^{2}\cos(\kappa)\sin^{3}(\kappa))\right)\right] \tag*{(61)}

the optimal ∣g2∣|g_2| then follows as

∣g2∣opt=1η4κ∣g1∣e−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)]1η+2∣g1∣2(1−e−8κ2σmeff2+8∣g1∣2sin⁡4(κ)cos⁡(2κ−8∣g1∣2cos⁡(κ)sin⁡3(κ)))(62)|g_2|_{\mathrm{opt}}=\frac{1}{\sqrt{\eta}}\frac{4\kappa|g_1|e^{-2\kappa^2\sigma_{m_{\mathrm{eff}}}^2}\left[\sigma_{m_0}^2+|g_1|^2\operatorname{sinc}(\kappa)\cos(\kappa)\right]}{\frac{1}{\eta}+2|g_1|^2\left(1-e^{-8\kappa^2\sigma_{m_{\mathrm{eff}}}^2+8|g_1|^2\sin^4(\kappa)}\cos\left(2\kappa-8|g_1|^2\cos(\kappa)\sin^3(\kappa)\right)\right)} \tag*{(62)}

and as before

σmf2=σm02+∣g1∣2−4κ∣g1∣η∣g2∣opte−2κ2σmeff2[σm02+∣g1∣2sinc⁡(κ)cos⁡(κ)](63)\sigma_{m_f}^2=\sigma_{m_0}^2+|g_1|^2-4\kappa|g_1|\sqrt{\eta}|g_2|_{\mathrm{opt}}e^{-2\kappa^2\sigma_{m_{\mathrm{eff}}}^2}\left[\sigma_{m_0}^2+|g_1|^2\operatorname{sinc}(\kappa)\cos(\kappa)\right] \tag*{(63)}

As before, for κ≪1\kappa\ll1,

η∣g2∣opt≃4κσmeff2∣g1∣e−2κ2σmeff21η+2∣g1∣2(1−e−8κ2σmeff2)(64)\sqrt{\eta}|g_2|_{\mathrm{opt}}\simeq\frac{4\kappa\sigma_{m_{\mathrm{eff}}}^2|g_1|e^{-2\kappa^2\sigma_{m_{\mathrm{eff}}}^2}}{\frac{1}{\eta}+2|g_1|^2\left(1-e^{-8\kappa^2\sigma_{m_{\mathrm{eff}}}^2}\right)} \tag*{(64)}

and

σmf2(∣g2∣opt)≃σmeff2[1−16κ2σmeff2η∣g1∣2cosh⁡(4κ2σmeff2)+(1+4η∣g1∣2)sinh⁡(4κ2σmeff2)](65)\sigma_{m_f}^2\left(|g_2|_{\mathrm{opt}}\right)\simeq\sigma_{m_{\mathrm{eff}}}^2\left[1-\frac{16\kappa^2\sigma_{m_{\mathrm{eff}}}^2\eta|g_1|^2}{\cosh\left(4\kappa^2\sigma_{m_{\mathrm{eff}}}^2\right)+\left(1+4\eta|g_1|^2\right)\sinh\left(4\kappa^2\sigma_{m_{\mathrm{eff}}}^2\right)}\right] \tag*{(65)}

At this point, this is identical to before with the replacement ∣g1∣2→η∣g1∣2|g_1|^2\rightarrow\eta|g_1|^2, though not in σmeff2\sigma_{m_{\mathrm{eff}}}^2, which makes sense, as that noise is induced before any loss. Note also that this immediately gives a precise limit on transmission to achieve cooling, given by the point where σmf2=σm02\sigma_{m_f}^2=\sigma_{m_0}^2:

ηmin⁡=e4κ2σmeff216κ2σmeff4−4∣g1∣2sinh⁡(4κ2σmeff2)(66)\eta_{\min}=\frac{e^{4\kappa^2\sigma_{m_{\mathrm{eff}}}^2}}{16\kappa^2\sigma_{m_{\mathrm{eff}}}^4-4|g_1|^2\sinh\left(4\kappa^2\sigma_{m_{\mathrm{eff}}}^2\right)} \tag*{(66)}

The same optimum follows from before as long as 1+4η∣g1∣21+4\eta|g_1|^2 remains reasonably large:

κopt≃12σmeff(31+4η∣g1∣2)1/6(67)\kappa_{\mathrm{opt}}\simeq\frac{1}{2\sigma_{m_{\mathrm{eff}}}}\left(\frac{3}{1+4\eta|g_1|^2}\right)^{1/6} \tag*{(67)}

Now the asymptotic energy spread follows as

σmf2≃32/3σm02+∣g1∣24×21/3η2/3∣g1∣4/3(68)\sigma_{m_f}^2\simeq\frac{3^{2/3}\sigma_{m_0}^2+|g_1|^2}{4\times2^{1/3}\eta^{2/3}|g_1|^{4/3}} \tag*{(68)}

The optimal value of the pickup coupling is unaffected, it is still ∣g1∣=2σm0|g_1|=\sqrt{2}\sigma_{m_0}, only now the optimal energy spread is σmf=35/38η−1/3σm01/3\sigma_{m_f}=\sqrt{\frac{3^{5/3}}{8}}\eta^{-1/3}\sigma_{m_0}^{1/3}.

Figure 6 explores the effect of photon loss between the pickup and kicker on the same case study from the main text. Panel (a) shows the spectrum as a function of transmission, reflecting both the reduction in effective kicker coupling to ηg2\sqrt{\eta}g_2 and the added noise from decoherence into the loss mode, and with no direct correction for the loss. Panel (b) shows the spectrum now using the corrected optimal values for ∣g2∣|g_2| and κ\kappa. It is clear that the spectrum remains narrower down to lower pickup-kicker transmissions when using the corrected optimal values, and the net energy decreases further as the optimal g2g_2 grows like 1/η1/\sqrt{\eta}. Panel (c) shows the final energy spread and state purity as a function of the transmission for the uncorrected (solid) and corrected (dashed) cases. For the energy spread, we also plot the asymptotic energy spread estimate scaling as η−1/3\eta^{-1/3}. Using loss-corrected optimal values for the drift length and kicker coupling maintains good cooling performance even down to around 30% transmission.

Figure 6. Study of loss effects on cooling performance. Panels (a) and (b) show the uncorrected and corrected spectra as a function of the pickup-to-kicker transmission η\eta. Panel (c) summarizes the final energy spread and state purity as a function of the transmission.

References

  1. [1]P. Musumeci, J. G. Navarro, J. Rosenzweig, L. Cultrera, I. Bazarov, J. Maxson, S. Karkare, and H. Padmore, Advances in bright electron sources, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 907, 209 (2018).DOI
  2. [2]J. C. Spence, High-resolution electron microscopy (OUP Oxford, 2013).
  3. [3]P. E. Batson, N. Dellby, and O. L. Krivanek, Sub-ångstrom resolution using aberration corrected electron optics, Nature 418, 617 (2002).pubmed.ncbi.nlm.nih.gov/12167855
  4. [4]O. L. Krivanek, T. C. Lovejoy, N. Dellby, T. Aoki, P. Rez, E. Soignard, J. Zhu, P. E. Batson, M. J. Lagos, et al., Vibrational spectroscopy in the electron microscope, Nature 514, 209 (2014).
  5. [5]A. H. Zewail, Four-dimensional electron microscopy, Science 328, 187 (2010).DOI
  6. [6]A. Feist, N. Bach, N. R. da Silva, T. Danz, M. Möller, K. E. Priebe, T. Domröse, J. G. Gatzmann, S. Rost, J. Schausss, et al., Ultrafast transmission electron microscopy using a laser-driven field emitter: Femtosecond resolution with a high coherence electron beam, Ultramicroscopy 176, 63 (2017).DOI
  7. [7]A. Arbouet, G. M. Caruso, and F. Houdellier, Ultrafast transmission electron microscopy: historical development, instrumentation, and applications, Advances in imaging and electron physics 207, 1 (2018).DOI
  8. [8]F. Houdellier, G. M. Caruso, S. Weber, M. Kociak, and A. Arbouet, Development of a high brightness ultrafast transmission electron microscope based on a laser-driven cold field emission source, Ultramicroscopy 186, 128 (2018).
  9. [9]A. Schröder, A. Wendel, J. T. Weber, M. Mukai, Y. Kohno, and S. Schäfer, Laser-driven cold field emission source for ultrafast transmission electron microscopy, Ultramicroscopy 275, 114158 (2025).
  10. [10]T. Chlouba, M. Solà-Garcia, N. van Nielen, M. Wilts-Liebtrau, J. Litzel, L. Brückner, R. Shiloh, E. Kieft, P. Kruit, P. Hommelhoff, et al., Controlling charge, energy spread, and pulse duration in an ultrafast scanning electron microscope, Ultramicroscopy , 114462 (2026).
  11. [11]J. Spence, W. Qian, and M. Silverman, Electron source brightness and degeneracy from fresnel fringes in field emission point projection microscopy, Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films 12, 542 (1994).
  12. [12]H. Kiesel, A. Renz, and F. Hasselbach, Observation of hanbury brown–twiss anticorrelations for free electrons, Nature 418, 392 (2002).pubmed.ncbi.nlm.nih.gov/12140550
  13. [13]M. Kuwahara, S. Kusunoki, Y. Nambo, K. Saitoh, X. Jin, T. Ujihara, H. Asano, Y. Takeda, and N. Tanaka, Coherence of a spin-polarized electron beam emitted from a semiconductor photocathode in a transmission electron microscope, Applied Physics Letters 105 (2014).DOI
  14. [14]K. Tsuno, T. Kaneyama, T. Honda, K. Tsuda, M. Terauchi, and M. Tanaka, Design and testing of omega mode imaging energy filters at 200 kv, Microscopy 46, 357 (1997).DOI
  15. [15]O. L. Krivanek, N. Dellby, J. A. Hachtel, J.-C. Idrobo, M. T. Hotz, B. Plotkin-Swing, N. J. Bacon, A. L. Bleloch, G. J. Corbin, M. V. Hoffman, et al., Progress in ultrahigh energy resolution eels, Ultramicroscopy 203, 60 (2019).DOI
  16. [16]C. Kealhofer, W. Schneider, D. Ehberger, A. Ryabov, F. Krausz, and P. Baum, All-optical control and metrology of electron pulses, Science 352, 429 (2016).DOI
  17. [17]M. Yannai, Y. Adiv, R. Dahan, K. Wang, A. Gorlach, N. Rivera, T. Fishman, M. Krüger, and I. Kaminer, Lossless monochromator in an ultrafast electron microscope using near-field thz radiation, Physical Review Letters 131, 145002 (2023).DOI
  18. [18]N. L. Streshkova, P. Koutenský, T. Novotný, and M. Kozák, Monochromatization of electron beams with spatially and temporally modulated optical fields, Physical Review Letters 133, 213801 (2024).DOI
  19. [19]S. Van der Meer, Stochastic damping of betatron oscillations, CERN ISR-PO 72, 31 (1972).
  20. [20]D. Möhl, G. Petrucci, L. Thorndahl, and S. Van Der Meer, Physics and technique of stochastic cooling, Physics Reports 58, 73 (1980).DOI
  21. [21]S. Van Der Meer, Stochastic cooling and the accumulation of antiprotons, Reviews of Modern Physics 57, 689 (1985).
  22. [22]A. Mikhailichenko and M. Zolotorev, Optical stochastic cooling, Physical Review Letters 71, 4146 (1993).pubmed.ncbi.nlm.nih.gov/10055168
  23. [23]M. Zolotorev and A. Zholents, Transit-time method of optical stochastic cooling, Physical Review E 50, 3087 (1994).
  24. [24]V. Lebedev, J. Jarvis, H. Piekarz, A. Romanov, J. Ruan, and M. Andorf, The design of optical stochastic cooling for iota, Journal of Instrumentation 16 (05), T05002.
  25. [25]J. Jarvis, V. Lebedev, A. Romanov, D. Broemmelsiek, K. Carlson, S. Chattopadhyay, A. Dick, D. Edstrom, I. Lobach, S. Nagaitsev, et al., Experimental demonstration of optical stochastic cooling, Nature 608, 287 (2022).DOI
  26. [26]C. Pellegrini, A. Marinelli, and S. Reiche, The physics of x-ray free-electron lasers, Reviews of Modern Physics 88, 015006 (2016).DOI
  27. [27]F. J. Garcia de Abajo, Optical excitations in electron microscopy, Reviews of modern physics 82, 209 (2010).DOI
  28. [28]J. Breuer and P. Hommelhoff, Laser-based acceleration of nonrelativistic electrons at a dielectric structure, Physical review letters 111, 134803 (2013).pubmed.ncbi.nlm.nih.gov/24116785
  29. [29]E. Peralta, K. Soong, R. England, E. Colby, Z. Wu, B. Montazeri, C. McGuinness, J. McNeur, K. Leedle, D. Walz, et al., Demonstration of electron acceleration in a laser-driven dielectric microstructure, Nature 503, 91 (2013).DOI
  30. [30]Y. Yang, A. Massuda, C. Roques-Carmes, S. E. Kooi, T. Christensen, S. G. Johnson, J. D. Joannopoulos, O. D. Miller, I. Kaminer, and M. Soljačić, Maximal spontaneous photon emission and energy loss from free electrons, Nature Physics 14, 894 (2018).
  31. [31]O. Kfir, Entanglements of electrons and cavity photons in the strong-coupling regime, Physical review letters 123, 103602 (2019).DOI
  32. [32]O. Kfir, H. Lourenço-Martins, G. Storeck, M. Sivis, T. R. Harvey, T. J. Kippenberg, A. Feist, and C. Ropers, Controlling free electrons with optical whispering-gallery modes, Nature 582, 46 (2020).
  33. [33]R. Dahan, A. Gorlach, U. Haeusler, A. Karnieli, O. Eyal, P. Yousefi, M. Segev, A. Arie, G. Eisenstein, P. Hommelhoff, et al., Imprinting the quantum statistics of photons on free electrons, Science 373, eabj7128 (2021).DOI
  34. [34]A. Feist, G. Huang, G. Arend, Y. Yang, J.-W. Henke, A. S. Raja, F. J. Kappert, R. N. Wang, H. Lourenço-Martins, Z. Qiu, et al., Cavity-mediated electron-photon pairs, Science 377, 777 (2022).DOI
  35. [35]Y. Adiv, H. Hu, S. Tesses, R. Dahan, K. Wang, Y. Kurman, A. Gorlach, H. Chen, X. Lin, G. Bartal, et al., Observation of 2d cherenkov radiation, Physical Review X 13, 011002 (2023).DOI
  36. [36]C. Roques-Carmes, S. E. Kooi, Y. Yang, N. Rivera, P. D. Keathley, J. D. Joannopoulos, S. G. Johnson, I. Kaminer, K. K. Berggren, and M. Soljačić, Free-electron–light interactions in nanophotonics, Applied Physics Reviews 10 (2023).
  37. [37]T. Chlouba, R. Shiloh, S. Kraus, L. Brückner, J. Litzel, and P. Hommelhoff, Coherent nanophotonic electron accelerator, Nature 622, 476 (2023).DOI
  38. [38]V. Di Giulio, E. Akerboom, A. Polman, and F. J. García de Abajo, Toward optimum coupling between free electrons and confined optical modes, ACS nano 18, 14255 (2024).DOI
  39. [39]F. J. G. De Abajo, A. Polman, C. I. Velasco, M. Kociak, L. H. Tizei, O. Stéphans, S. Meuret, T. Sannomiya, K. Akiba, Y. Auad, et al., Roadmap for quantum nanophotonics with free electrons, ACS photonics 12, 4760 (2025).
  40. [40]Z. Xie, Z. Chen, H. Li, Q. Yan, H. Chen, X. Lin, I. Kaminer, O. D. Miller, and Y. Yang, Maximal quantum interaction between free electrons and photons, Physical Review Letters 134, 043803 (2025).DOI
  41. [41]Z. Zhao, Upper bound for the quantum coupling between free electrons and photons, Physical Review Letters 134, 043804 (2025).
  42. [42]G. Arend, G. Huang, A. Feist, Y. Yang, J.-W. Henke, Z. Qiu, H. Jeng, A. S. Raja, R. Haindl, R. N. Wang, et al., Electrons herald non-classical light, Nature Physics 21, 1855 (2025).DOI
  43. [43]V. Di Giulio, M. Kociak, and F. J. G. de Abajo, Probing quantum optical excitations with fast electrons, Optica 6, 1524 (2019).
  44. [44]F. J. Garcia de Abajo and A. Konečná, Optical modulation of electron beams in free space, Physical Review Letters 126, 123901 (2021).DOI
  45. [45]R. Ruimy, A. Karnieli, and I. Kaminer, Free-electron quantum optics, Nature Physics 21, 193 (2025).DOI
  46. [46]S. Heifets and M. Zolotorev, Quantum theory of optical stochastic cooling, Physical Review E 65, 016507 (2001).
  47. [47]See Supplemental Material, which includes Refs. [43, 44, 64–71] for more detailed derivations.
  48. [48]Y. Fang, J. K. Kutt ruff, Z. Zhao, L. Moehrle, P. Hommelhoff, and P. Baum, Quantum tomography of free electrons, arXiv preprint arXiv:2606.25397 (2026).arxiv.org/abs/2606.25397
  49. [49]B. Barwick, D. J. Flannigan, and A. H. Zewail, Photon-induced near-field electron microscopy, Nature 462, 902 (2009).DOI
  50. [50]S. T. Park, M. Lin, and A. H. Zewail, Photon-induced near-field electron microscopy (pinem): theoretical and experimental, New Journal of Physics 12, 123028 (2010).DOI
  51. [51]A. Feist, K. E. Echternkamp, J. Schauss, S. V. Yalunin, S. Schäfer, and C. Ropers, Quantum coherent optical phase modulation in an ultrafast transmission electron microscope, Nature 521, 200 (2015).DOI
  52. [52]R. Shiloh, T. Chlouba, and P. Hommelhoff, Quantum-coherent light-electron interaction in a scanning electron microscope, Physical Review Letters 128, 253301 (2022).DOI
  53. [53]K. Wang and D. M. Mittleman, Metal wires for terahertz wave guiding, Nature 432, 376 (2004).pubmed.ncbi.nlm.nih.gov/15549101
  54. [54]J. Anthony, R. Leonhardt, S. G. Leon-Saval, and A. Argyros, Thz propagation in kagome hollow-core microstructured fibers, Optics express 19, 18470 (2011).DOI
  55. [55]A. Leitenstorfer, A. S. Moskalenko, T. Kampfrath, J. Kono, E. Castro-Camus, K. Peng, N. Qureshi, D. Turchinovich, K. Tanaka, A. G. Markelz, et al., The 2023 terahertz science and technology roadmap, Journal of Physics D: Applied Physics 56, 223001 (2023).
  56. [56]J. A. Harrington, A review of transmitting, hollow waveguides, Fiber & Integrated Optics 19, 211 (2000).DOI
  57. [57]B. Temelkuran, S. D. Hart, G. Benoit, J. D. Joannopoulos, and Y. Fink, Wavelength-scale hollow optical fibres with large photonic bandgaps for co2 laser transmission, Nature 420, 650 (2002).pubmed.ncbi.nlm.nih.gov/12478288
  58. [58]A. Ventura, J. G. Hayashi, J. Cimek, G. Jasion, P. Janicek, F. B. Slimen, N. White, Q. Fu, L. Xu, H. Sakr, et al., Extruded tellurite antiresonant hollow core fiber for mid-ir operation, Optics Express 28, 16542 (2020).DOI
  59. [59]P. Kruit, R. G. Hobbs, C.-S. Kim, Y. Yang, V. R. Manfrinato, J. Hammer, S. Thomas, P. Weber, B. Klopfer, C. Kohstall, et al., Designs for a quantum electron microscope, Ultramicroscopy 164, 31 (2016).arxiv.org/abs/1510.05946
  60. [60]T. Juffmann, S. A. Koppell, B. B. Klopfer, C. Ophus, R. M. Glaeser, and M. A. Kasevich, Multi-pass transmission electron microscopy, Scientific reports 7, 1699 (2017).DOI
  61. [61]S. A. Koppell, M. Mankos, A. J. Bowman, Y. Israel, T. Juffmann, B. B. Klopfer, and M. A. Kasivevich, Design for a 10 kev multi-pass transmission electron microscope, Ultramicroscopy 207, 112834 (2019).arxiv.org/abs/1904.11064
  62. [62]R. Zimmermann, M. Seidling, and P. Hommelhoff, Charged particle guiding and beam splitting with autoponderomotive potentials on a chip, Nature Communications 12, 390 (2021).
  63. [63]M. Seidling, F. Schmidt-Kaler, R. Zimmermann, J. Simonaitis, P. Keathley, K. Berggren, and P. Hommelhoff, Resonating electrostatically guided electrons, Physical Review Letters 132, 255001 (2024).DOI
  64. [64]C. M. Stein, Estimation of the mean of a multivariate normal distribution, The annals of Statistics , 1135 (1981).DOI
  65. [65]N. Talebi, Strong interaction of slow electrons with near-field light visited from first principles, Physical Review Letters 125, 080401 (2020).DOI
  66. [66]A. Karnieli and S. Fan, Jaynes-cummings interaction between low-energy free electrons and cavity photons, Science advances 9, eadh2425 (2023).arxiv.org/abs/2302.01575
  67. [67]A. Karnieli, C. Roques-Carmes, N. Rivera, and S. Fan, Strong coupling and single-photon nonlinearity in free-electron quantum optics, ACS Photonics 11, 3401 (2024).arxiv.org/abs/2403.13071
  68. [68]M. Sirotin, A. Rasputnyi, T. Chlouba, R. Shiloh, and P. Hommelhoff, Quantum computing and quantum optics with recoiled free electrons, arXiv preprint arXiv:2405.06560 (2024).DOI
  69. [69]A. P. Synaidis, P. Gonçalves, C. Ropers, and F. J. G. de Abajo, Quantum effects in the interaction of low-energy electrons with light, Science Advances 10, eadp4096 (2024).
  70. [70]U. Leonhardt and H. Paul, Measuring the quantum state of light, Progress in Quantum Electronics 19, 89 (1995).
  71. [71]U. Leonhardt, Measuring the quantum state of light, Vol. 22 (Cambridge university press, 1997).

Paper details

vlnc:W_ssg78syk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents