betaFeedback

WorksPhysics and Astronomy

Free-electron decoherence: Theory and applications

Abstract

Electron microscopy relies on the spatial coherence of electron beams to generate atomic-scale images using interference and diffraction, which can be degraded by inelastic scattering processes that induce decoherence. Here, we present a theoretical study of decoherence arising from the electromagnetic interaction of free electrons with bulk materials and planar surfaces. We show that bulk plasmons dominate decoherence in Al and Au, while electronic excitations above the band gap, supplemented by weaker coupling to phononic and guided modes, are the primary channels in ionic insulators such as LiF. A thermal population of electromagnetic modes leads to a divergence in the energy-loss probability at low frequencies, which in turn produces a pronounced temperature dependence. We show that this effect can be exploited for nanoscale thermometry, predicting that optimized energy-filtered holography enables ∼0.1%\sim0.1\% changes in fringe visibility for physically viable temperature variations in metals. Through these results, we establish a unified theoretical framework to describe free-electron decoherence in the bulk and surfaces of arbitrary materials.

Introduction

In electron microscopy, free electrons serve as unique nanoscale probes thanks to their evanescent broadband electromagnetic fields and short de Broglie wavelength (λe=2−39\lambda_e = 2 - 39 pm for kinetic energies in the 300 - 1 keV range), which pushes the diffraction limit well below the size of an atom. These properties allow free electrons to couple to modes that are inaccessible to free-space radiation while also enabling atomic-scale spatial resolution. In addition, electron microscopes currently implement techniques such as low-energy [1, 2] and ultrafast [3, 4] electron diffraction, as well as holography [5, 6], which can be performed using laterally extended electron beams (e-beams) and exploit the wave-like nature of free electrons to image the phase produced by elastic scattering by the atomic lattice of a specimen. In this context, the inelastic scattering associated with the exchange of energy quanta between the electron and various types of excitations (e.g., electronic [7, 8, 9, 10, 11, 12, 13, 14, 15], radiative [16, 17, 18], and surface modes in general [8, 10, 16, 19]) can undermine this working principle by introducing decoherence between spatially separated components.

Free-electron decoherence is produced when separated spatial components of the electron wavefunction couple to the environment with different amplitudes, thereby generating which-path information. Because incoherent components cannot interfere, they contribute only a uniform background to the interference pattern, reducing the fringe contrast and degrading the visibility of interference fringes in electron holography. However, interference from inelastically scattered electrons remains possible when different lateral positions of the incident wavefunction interact similarly with the extended modes of a specimen. In particular, different lateral positions of energy-filtered electrons that have excited plasmons while traversing a metallic film can still interfere if the propagation length of those plasmons is larger than the separation between those positions. This is the so-called inelastic electron holography, which has been demonstrated in several experiments [20, 21, 22, 23, 24]. Understanding of the different sources of decoherence thus becomes an important goal.

Earlier theoretical studies focused on the exploration of decoherence in two-path electron interferometry near planar surfaces [7, 8, 9, 10, 11, 25, 26], followed by experimental demonstrations [12, 13, 14, 15, 27]. In these works, the dominant contributions to decoherence arise from coupling to electromagnetic modes propagating along the material surface with phase velocities matching the electron speed. More recently, decoherence has also been predicted in situations where electrons pass near the edge of extended structures, resulting in the emission of free radiation [18]. In this regime, temperature-induced spontaneous emission and absorption of low-energy radiative modes drastically increase the loss of coherence. Since the electron interacts only with radiation fields without exciting internal modes of the sample, this mechanism has been proposed as a route for noncontact detection of distant objects [18].

Free-electron decoherence can be studied by analyzing the real-space representation of the reduced density matrix ρ(r,r′)\rho(\mathbf{r},\mathbf{r}'), obtained from the entire joint electron-environment quantum state by tracing out all degrees of freedom associated with the environment (including radiation and material excitations), assumed to be in thermal equilibrium at a temperature TT. Starting with an electron moving along the zz axis and entering the interaction region in a state ρi(r,r′)\rho_i(\mathbf{r},\mathbf{r}') (see schematic in Figure 1a), the post-interaction density matrix becomes (see Sec. II below and Refs. [18, 19]

Figure 1. Free-electron decoherence and interferometry. (a) An incident electron prepared in an arbitrary state corresponding to a density matrix ρi(r,r′)\rho_i(\mathbf{r},\mathbf{r}') interacts with a specimen and evolves according to Eq. (1), acquiring an elastic phase χ\chi and undergoing a loss of coherence (probability PP) associated with inelastic processes. (b) We consider an electron prepared in a two-path superposition (interpath distance d⊥d_{\perp}, Gaussian path radius w0w_0) such that each path excites the specimen with a different amplitude. Electron paths are then recombined at a plane zoz_o upon reflection on electrostatic mirrors and propagation over a distance d∥d_{\parallel}. A microscope consisting of two lenses with focal lengths f1f_1 and f2f_2 maps the zoz_o plane onto an enlarged image at a plane ziz_i, where an electron detector records the magnified interference fringes. The distance d1d_1 from zoz_o to lens 1 determines the distance d2d_2 from ziz_i to lens 2 (see Appendix A), while the magnification factor is fixed by g=f2/f1g=f_2/f_1. (c) Interference patterns formed with 200 keV electrons under the configuration of panel (b) as a function of lateral position xx for g=5000g=5000, d∥/d⊥=104d_{\parallel}/d_{\perp}=10^4, w0=200w_0=200 nm, and two different values of the interpath decoherence probability PABP_{AB} with χAB=0\chi_{AB}=0.

ρ(r,r′)=e−P(r,r′)+iχ(r,r′)ρi(r,r′).(1)\rho(\mathbf{r},\mathbf{r}')=e^{-P(\mathbf{r},\mathbf{r}')+i\chi(\mathbf{r},\mathbf{r}')}\rho_i(\mathbf{r},\mathbf{r}'). \tag*{(1)}

Here, P(r,r′)P(\mathbf{r},\mathbf{r}') acts as a decoherence probability accounting for an exponential loss of coherence between the transverse spatial components of ρi\rho_i, encapsulating all the inelastic interactions undergone by the electron. Also, χ(r,r′)\chi(\mathbf{r},\mathbf{r}') embodies an elastic phase shift associated with polarization of the environment [19].

In this work, we focus on P(r,r′)P(\mathbf{r},\mathbf{r}') and provide a detailed exploration of electron coupling to planar structures made of different materials. We analyze the inelastic processes that lead to decoherence, highlighting the distinction between bulk, surface, and radiative excitations. The article is organized as follows: In Sec. II, we present a general theoretical framework to compute the temperature-dependent decoherence for any arbitrary material and geometry, described in terms of the electromagnetic Green tensor. Additionally, we propose a simple experimental configuration as an example to measure both PP and χ\chi. In § 1.2 and § 1.3, we analyze different sources of decoherence, and in particular, in § 1.2, we study bulk-induced decoherence for an e-beam split into different paths propagating inside an infinitely extended medium made of either Au, Al, or LiF. In § 1.3, we examine surface effects associated with configurations in which two paths travel either parallel or perpendicular to planar surfaces. In § 1.4, we study decoherence as a function of temperature in the specimen, exploring the temperature sensitivity achieved by measuring the fringe visibility in energy-filtered holography. We conclude in § 2 by summarizing our findings and discussing the potential of decoherence as a novel probe for thermal sensing, compared with other existing techniques.

Free-electron spatial decoherence

Effect of decoherence on electron interference

To study how decoherence influences the evolution of free electrons after the interaction with material structures and scattered electromagnetic fields, we first discuss a simple tutorial scenario that relates closely to electron microscope setups. We consider the configuration sketched in Figure 1b, starting with an e-beam of relativistic electrons prepared in a two-path superposition under paraxial propagation conditions. We assume loosely focused Gaussian sub-beams, such that their transverse profile is preserved without diffraction during propagation along the entire length of the setup (e.g., over the Rayleigh range zR=πw02/λe≈50z_R=\pi w_0^2/\lambda_e\approx50 nm for 200 keV electrons and a Gaussian radius w0=200w_0=200 nm). In addition, we assume a small path width compared to the characteristic spatial variation of elastic and inelastic interactions. One of the paths passes through or close to a specimen, where it acquires some decoherence relative to the other path. In general, interpath decoherence arises when there is a different degree of interaction between each path and the specimen. After interaction, the paths are reflected by electrostatic mirrors with a small deflection angle that preserves paraxial propagation. The two paths overlap after a propagation distance d∥d_{\parallel}, where they display interference fringes. Finally, the overlapping region (object plane zoz_{\mathrm{o}}) is magnified by a factor f2/f1f_2/f_1 upon transmission through a microscope consisting of two lenses of focal lengths f1f_1 and f2f_2, separated by a distance f1+f2f_1+f_2 (see Figure 1b). An electron detector is placed at the image plane ziz_{\mathrm{i}}.

For simplicity, we consider an interpath separation d⊥≫w0d_{\perp} \gg w_0. As anticipated in Eq. (1), the two electron paths, which we denote A and B, acquire a relative decoherence probability PAB=PBAP_{AB}=P_{BA}, which reduces the visibility of the interference pattern, accompanied by an elastic phase shift χAB=−χBA\chi_{AB}=-\chi_{BA}. The latter produces a lateral shift of the fringes. Furthermore, we neglect intrapath decoherence and phase shifting, assuming that the path radius w0w_0 is small compared with the spatial frequencies of the electromagnetic fields mediating the electron interaction with the specimen. Under these conditions, the intensity recorded in the image plane as a function of lateral position xx reduces to (see detailed derivation in Appendix A)

I(x)∝e−2x2/wi2[1+e−PABcos⁡(x/di+χAB)],(2)I(x) \propto e^{-2x^2/w_i^2}\left[1+e^{-P_{AB}}\cos\left(x/d_i+\chi_{AB}\right)\right], \tag*{(2)}

where the distances wi=gw0w_i=gw_0 and di=(g/2π)(d∥/d⊥)λed_i=(g/2\pi)(d_{\parallel}/d_{\perp})\lambda_{\mathrm{e}} are magnified by the lens factor g=f2/f1g=f_2/f_1 and the inverse numerical aperture d∥/d⊥d_{\parallel}/d_{\perp}.

The contrast of the interference pattern can be quantitatively characterized using the fringe visibility, defined as V=(Imax⁡−Imin⁡)/(Imax⁡+Imin⁡)V=(I_{\max}-I_{\min})/(I_{\max}+I_{\min}), which measures the normalized difference between the intensities at adjacent maxima and minima. From Eq. (2), we find V=e−PABV=e^{-P_{AB}} at the center of the interference pattern (assuming wi≫diw_i\gg d_i). Consequently, increasing the decoherence probability PABP_{AB} leads to an exponential suppression of the contrast of the interference pattern. While this loss of contrast is usually interpreted as a degradation of the information accessible through interferometric measurements, it can also be exploited as a probe of environmental properties. In particular, since PABP_{AB} depends on the sample temperature TT, measurements of the visibility can provide a means of inferring local thermal conditions, as discussed in § 1.4.

General theory of electron decoherence

To obtain a general explicit form of Eq. (1), we represent the electron state through the real-space density matrix ρ(r,r′)\rho(\mathbf{r},\mathbf{r}') and study its evolution during the interaction with materials and scattered electromagnetic fields. We work under the nonrecoil approximation, assuming that the initial momentum uncertainty and the momentum exchanges associated with the interaction are both negligible compared to the central electron momentum ℏq0\hbar\mathbf{q}_0. The latter is taken along the zz direction without loss of generality. We define the central electron kinetic energy E0=mec2[1+(ℏq0/mec)2−1]E_0=m_{\mathrm{e}}c^2\left[\sqrt{1+(\hbar q_0/m_{\mathrm{e}}c)^2}-1\right] and velocity

v=c1−1/(E0/mec2+1)2v=c\sqrt{1-1/(E_0/m_{\mathrm{e}}c^2+1)^2}, as well as the Lorentz factor γ=1/1−v2/c2\gamma=1/\sqrt{1-v^2/c^2}.

We consider the system formed by the electron (el) and its environment (env). The latter consists of material structures and radiation fields, including polaritons and other excitations. The total Hamiltonian of the system can be written in real-space electron coordinates as H^(r)=H^el(r)+H^env+H^int(r)\hat{\mathcal{H}}(\mathbf{r})=\hat{\mathcal{H}}_{\mathrm{el}}(\mathbf{r})+\hat{\mathcal{H}}_{\mathrm{env}}+\hat{\mathcal{H}}_{\mathrm{int}}(\mathbf{r}), which comprises the noninteracting electron and environment components H^el(r)\hat{\mathcal{H}}_{\mathrm{el}}(\mathbf{r}) and H^env\hat{\mathcal{H}}_{\mathrm{env}}, as well as the interaction term H^int(r)\hat{\mathcal{H}}_{\mathrm{int}}(\mathbf{r}). By linearizing the electron energy Eq≈E0+ℏv⋅(q−q0)E_{\mathbf{q}}\approx E_0+\hbar\mathbf{v}\cdot(\mathbf{q}-\mathbf{q}_0) as a function of wavevector q\mathbf{q} relative to q0\mathbf{q}_0, we can approximate the electron Hamiltonian as H^el(r)=E0−iℏv⋅(q0+i∇)\hat{\mathcal{H}}_{\mathrm{el}}(\mathbf{r})=E_0-i\hbar\mathbf{v}\cdot(\mathbf{q}_0+i\nabla). The environment consists of electromagnetic modes of energies ℏωi\hbar\omega_i, which include their interaction with the materials. We write the corresponding Hamiltonian as H^env=ℏ∑iωia^i†a^i\hat{\mathcal{H}}_{\mathrm{env}}=\hbar\sum_i\omega_i\hat{a}_i^\dagger\hat{a}_i, where a^i†\hat{a}_i^\dagger and a^i\hat{a}_i are creation and annihilation operators. We assume that all modes under consideration have a bosonic nature (i.e., [a^i,a^j†]=δij[\hat{a}_i,\hat{a}_j^\dagger]=\delta_{ij}). Because the electron velocity is constant in the nonrecoil approximation, the spatially integrated electron current reduces to eve\mathbf{v}, and therefore, adopting the minimal-coupling prescription, orking in the temporal gauge (scalar potential set to zero), and neglecting A2A^2 terms, the interaction Hamiltonian becomes H^int(r)=(ev/c)⋅A^(r)\hat{\mathcal{H}}_{\mathrm{int}}(\mathbf{r})=(e\mathbf{v}/c)\cdot\hat{\mathbf{A}}(\mathbf{r}), where v=vz^\mathbf{v}=v\hat{\mathbf{z}}, while A^(r)=−ic∑iωi−1[Ei(r)a^i−Ei∗(r)a^i†]\hat{\mathbf{A}}(\mathbf{r})=-ic\sum_i\omega_i^{-1}[\boldsymbol{\mathcal{E}}_i(\mathbf{r})\hat{a}_i-\boldsymbol{\mathcal{E}}_i^*(\mathbf{r})\hat{a}_i^\dagger] is the vector potential operator constructed from the mode electric-field functions Ei(r)\boldsymbol{\mathcal{E}}_i(\mathbf{r}). We remark that the latter incorporates the interaction with material structures.

We switch to the interaction picture, where A^(r)\hat{\mathbf{A}}(\mathbf{r}) becomes time-dependent due to both the unitary transformation eiH^envt/ℏa^ie−iH^envt/ℏ=e−iωita^ie^{i\hat{\mathcal{H}}_{\mathrm{env}}t/\hbar}\hat{a}_i e^{-i\hat{\mathcal{H}}_{\mathrm{env}}t/\hbar}=e^{-i\omega_i t}\hat{a}_i and the spatial displacement induced by the −iℏv⋅∇-i\hbar\mathbf{v}\cdot\nabla term in H^el(r)\hat{\mathcal{H}}_{\mathrm{el}}(\mathbf{r}). In this picture, the interaction Hamiltonian becomes H^int(r,t)=(ev/c)⋅A^(r+vt,t)\hat{\mathcal{H}}_{\mathrm{int}}(\mathbf{r},t)=(e\mathbf{v}/c)\cdot\hat{\mathbf{A}}(\mathbf{r}+\mathbf{v}t,t), while the evolution of the electron–environment system is governed by the evolution operator S^(r,t)\hat{S}(\mathbf{r},t), relating the state at time tt to the initial state according to ∣ψ(r,t)⟩=S^(r,t)∣ψ(r,−∞)⟩|\psi(\mathbf{r},t)\rangle=\hat{S}(\mathbf{r},t)|\psi(\mathbf{r},-\infty)\rangle. Assuming that the electron and environment are initially uncorrelated, the initial state of the system is described by the product of the density matrix of the incident electron ρi(r,r′)\rho_i(\mathbf{r},\mathbf{r}') and a thermal environment state at a temperature TT. After the interaction (t→∞t\to\infty), we trace out the environment to obtain the reduced final density matrix of the electron

ρf(r,r′)=⟨S^(r,∞)S^†(r′,∞)⟩Tρi(r,r′),(3)\rho_f(\mathbf{r},\mathbf{r}')=\left\langle\hat{S}(\mathbf{r},\infty)\hat{S}^{\dagger}(\mathbf{r}',\infty)\right\rangle_T\rho_i(\mathbf{r},\mathbf{r}'), \tag*{(3)}

where ⟨⋯ ⟩T\langle\cdots\rangle_T denotes the thermal average over the environmental degrees of freedom. Finally, we recast Eq. (3) into the exponential form of Eq. (1) by identifying −P(r,r′)-P(\mathbf{r},\mathbf{r}') and χ(r,r′)\chi(\mathbf{r},\mathbf{r}') as the real and imaginary parts of log⁡⟨S^(r,∞)S^†(r′,∞)⟩T\log\langle\hat{S}(\mathbf{r},\infty)\hat{S}^{\dagger}(\mathbf{r}',\infty)\rangle_T, respectively.

Using the Magnus expansion [28], the evolution operature can be written as

S^(r,∞)=exp⁡{−iℏ∫−∞∞dt H^int(r,t)−12ℏ2∫−∞∞dt∫−∞tdt′ [H^int(r,t),H^int(r,t′)]}.(4)\hat{S}(\mathbf{r},\infty)=\exp\left\{-\frac{i}{\hbar}\int_{-\infty}^{\infty}dt\,\hat{H}_{\mathrm{int}}(\mathbf{r},t)-\frac{1}{2\hbar^2}\int_{-\infty}^{\infty}dt\int_{-\infty}^{t}dt'\,[\hat{H}_{\mathrm{int}}(\mathbf{r},t),\hat{H}_{\mathrm{int}}(\mathbf{r},t')]\right\}. \tag*{(4)}

It is important to stress that this expansion only produces contributions up to second order in the interaction Hamiltonian. Indeed, Eq. (4) is exact because the vector potential operator is linear in the ladder operators, and therefore, the commutator [A^z(r,t),A^z(r′,t′)][\hat{A}_z(\mathbf{r},t),\hat{A}_z(\mathbf{r}',t')] is a (purely imaginary) c-number, causing higher-order terms in the expansion to vanish.

Using the Baker–Campbell–Hausdorff formula, the product of the two evolution operators in Eq. (3) can be combined into a single exponential eC^(r,r′)+iχ(r,r′)e^{\hat{C}(\mathbf{r},\mathbf{r}')+i\chi(\mathbf{r},\mathbf{r}')}, where C^\hat{C} contains linear terms in the creation and annihilation operators, whereas χ\chi collects the scalar phases arising from the commutators. We evaluate the thermal average using the identity ⟨exp⁡{∑i(ci∗a^i†−cia^i)}⟩T=exp⁡{(1/2)∑i⟨(ci∗a^i†−cia^i)2⟩T}\langle\exp\{\sum_i(c_i^*\hat{a}_i^\dagger-c_i\hat{a}_i)\}\rangle_T=\exp\{(1/2)\sum_i\langle(c_i^*\hat{a}_i^\dagger-c_i\hat{a}_i)^2\rangle_T\}, which transforms Eqs. (3) and (4) into Eq. (1) with a decoherence probability and elastic phase given by

P(r,r′)=12(evℏc)2⟨{∫−∞∞dt [A^z(r−vt,t)−A^z(r′−vt,t)]}2⟩T,(5)P(\mathbf{r},\mathbf{r}')=\frac{1}{2}\left(\frac{ev}{\hbar c}\right)^2\left\langle\left\{\int_{-\infty}^{\infty}dt\,[\hat{A}_z(\mathbf{r}-\mathbf{v}t,t)-\hat{A}_z(\mathbf{r}'-\mathbf{v}t,t)]\right\}^2\right\rangle_T, \tag*{(5)}
χ(r,r′)=12(evℏc)2(∫−∞∞dt∫−∞∞dt′ [A^z(r−vt,t),A^z(r′−vt′,t′)]+∫−∞∞dt∫−∞tdt′{[A^z(r−vt,t),A^z(r−vt′,t′)]−[A^z(r′−vt,t),A^z(r′−vt′,t′)]}).(6)\chi(\mathbf{r},\mathbf{r}')=\frac{1}{2}\left(\frac{ev}{\hbar c}\right)^2\left(\int_{-\infty}^{\infty}dt\int_{-\infty}^{\infty}dt'\,[\hat{A}_z(\mathbf{r}-\mathbf{v}t,t),\hat{A}_z(\mathbf{r}'-\mathbf{v}t',t')] +\int_{-\infty}^{\infty}dt\int_{-\infty}^{t}dt'\{[\hat{A}_z(\mathbf{r}-\mathbf{v}t,t),\hat{A}_z(\mathbf{r}-\mathbf{v}t',t')]-[\hat{A}_z(\mathbf{r}'-\mathbf{v}t,t),\hat{A}_z(\mathbf{r}'-\mathbf{v}t',t')]\}\right). \tag*{(6)}

Both PP and χ\chi are contributed by A2A^2 terms, and we recall that higher-order terms vanish exactly for the bosonic excitations under consideration. Incidentally, the ponderomotive term that we have neglected in the interaction Hamiltonian could also produce an A2A^2 contribution through the linear term in Eq. (4), but with a relative weight ∼λe/L≪1\sim\lambda_e/L\ll1 compared to Eqs. (5), where LL is the characteristic interaction length determined by both geometry and the spatial extension of the excitations.

The thermal average in Eqs. (5) can be related to the macroscopic electromagnetic properties of the system by applying the fluctuation–dissipation theorem [18]

⟨A^z(r,t)A^z(r′,t′)⟩T=−4ℏc2∫0∞dω Im⁡{Gzz(r,r′,ω)}{2nT(ω)cos⁡[ω(t−t′)]+e−iω(t−t′)},(7)\langle\hat{A}_z(\mathbf{r},t)\hat{A}_z(\mathbf{r}',t')\rangle_T=-4\hbar c^2\int_0^\infty d\omega\,\operatorname{Im}\{G_{zz}(\mathbf{r},\mathbf{r}',\omega)\}\left\{2n_T(\omega)\cos[\omega(t-t')]+e^{-i\omega(t-t')}\right\}, \tag*{(7)}

where nT(ω)=(eℏω/kBT−1)−1n_T(\omega)=(e^{\hbar\omega/k_{\mathrm{B}}T}-1)^{-1} is the Bose–Einstein distribution at a temperature TT and Gzz(r,r′,ω)G_{zz}(\mathbf{r},\mathbf{r}',\omega) is the zzzz component of the 3×33\times3 electromagnetic Green tensor. The latter is implicitly defined by

∇×∇×G(r,r′,ω)−k2∫d3r′′ ϵ(r,r′′,ω)⋅G(r′′,r′,ω)=−1c2δ(r−r′)(8)\nabla\times\nabla\times\mathbf{G}(\mathbf{r},\mathbf{r}',\omega)-k^2\int d^3\mathbf{r}''\,\epsilon(\mathbf{r},\mathbf{r}'',\omega)\cdot\mathbf{G}(\mathbf{r}'',\mathbf{r}',\omega)=-\frac{1}{c^2}\delta(\mathbf{r}-\mathbf{r}') \tag*{(8)}

with k=ω/ck=\omega/c. Here, the response of the materials enters through the (generally nonlocal) permittivity tensor ϵ(r,r′,ω)\epsilon(\mathbf{r},\mathbf{r}',\omega). Note that, in Eq. (7), we have taken the continuum limit for the energies of the electromagnetic modes, transforming the discrete sum over ii into an integral over ω\omega [18] for a self-contained derivation of Eqs. (5) and (7). Assuming we are dealing with reciprocal materials, such that Gzz(r,r′,ω)=Gzz(r′,r,ω)G_{zz}(\mathbf{r},\mathbf{r}',\omega)=G_{zz}(\mathbf{r}',\mathbf{r},\omega), we can use Eq. (7) to readily obtain the commutator

[A^z(r,t),A^z(r′,t′)]=8iℏc2∫0∞dω sin⁡[ω(t−t′)]×Im⁡{Gzz(r,r′,ω)}.(9)[\hat{A}_z(\mathbf{r},t),\hat{A}_z(\mathbf{r}',t')]=8i\hbar c^2\int_0^\infty d\omega\,\sin[\omega(t-t')]\times\operatorname{Im}\{G_{zz}(\mathbf{r},\mathbf{r}',\omega)\}. \tag*{(9)}

In Eq. (9), we have omitted the thermal average on the left-hand side because the commutator is a c-number. This is corroborated by the fact that the temperature-dependent term in Eq. (7) cancels out when subtracting the two terms of the commutator.

Direct application of Eqs. (7) and (9) allows us to rewrite Eqs. (5) in a more convenient form in terms of the electromagnetic Green tensor:

P(r,r′)=∫0∞dω[nT(ω)+12][Γ(r,r,ω)+Γ(r′,r′,ω)−2Γ(r,r′,ω)](10)P(\mathbf{r},\mathbf{r}')=\int_{0}^{\infty}d\omega\left[n_T(\omega)+\frac{1}{2}\right]\left[\Gamma(\mathbf{r},\mathbf{r},\omega)+\Gamma(\mathbf{r}',\mathbf{r}',\omega)-2\Gamma(\mathbf{r},\mathbf{r}',\omega)\right] \tag*{(10)}
χ(r,r′)=2e2ℏ∫−∞∞dz′′∫−∞∞dz′′′∫0∞dω{2sin⁡[ωv(z−z′−z′′+z′′′)]Im⁡{Gzz(R,z′′,R′,z′′′,ω)}+cos⁡[ωv(z′′−z′′′)]Re⁡{Gzz(R′,z′′,R′,z′′′,ω)−Gzz(R,z′′,R,z′′′,ω)}}(11)\chi(\mathbf{r},\mathbf{r}')=\frac{2e^2}{\hbar}\int_{-\infty}^{\infty}dz''\int_{-\infty}^{\infty}dz'''\int_{0}^{\infty}d\omega\left\{2\sin\left[\frac{\omega}{v}(z-z'-z''+z''')\right]\operatorname{Im}\left\{G_{zz}(\mathbf{R},z'',\mathbf{R}',z''',\omega)\right\}\right. \\ \left.+\cos\left[\frac{\omega}{v}(z''-z''')\right]\operatorname{Re}\left\{G_{zz}(\mathbf{R}',z'',\mathbf{R}',z''',\omega)-G_{zz}(\mathbf{R},z'',\mathbf{R},z''',\omega)\right\}\right\} \tag*{(11)}

where

Γ(r,r′,ω)=4e2ℏ∫−∞∞dz′′∫−∞∞dz′′′cos⁡[ωv(z−z′−z′′+z′′′)]Im⁡{−Gzz(R,z′′,R′,z′′′,ω)}(12)\Gamma(\mathbf{r},\mathbf{r}',\omega)=\frac{4e^2}{\hbar}\int_{-\infty}^{\infty}dz''\int_{-\infty}^{\infty}dz'''\cos\left[\frac{\omega}{v}(z-z'-z''+z''')\right]\operatorname{Im}\left\{-G_{zz}(\mathbf{R},z'',\mathbf{R}',z''',\omega)\right\} \tag*{(12)}

acts as a nonlocal electron energy-loss probability. Here, R=(x,y)\mathbf{R}=(x,y) is the transverse component of r\mathbf{r} relative to the e-beam.

From Eqs. (9), the decoherence probability is found to be a symmetric function of spatial positions [P(r,r′)=P(r′,r)][P(\mathbf{r},\mathbf{r}')=P(\mathbf{r}',\mathbf{r})], while the phase is antisymmetric [χ(r,r′)=−χ(r′,r)][\chi(\mathbf{r},\mathbf{r}')=-\chi(\mathbf{r}',\mathbf{r})]. For r=r′\mathbf{r}=\mathbf{r}', we have χ(r,r)=0\chi(\mathbf{r},\mathbf{r})=0, while Γ(r,r,ω)\Gamma(\mathbf{r},\mathbf{r},\omega) coincides with the loss probability in electron energy-loss spectroscopy (EELS) [29]. In addition, we also have P(r,r)=0P(\mathbf{r},\mathbf{r})=0, which implies that the interaction does not modify the diagonal of the density matrix.

In the nonrecoil approximation here employed, Γ(r,r′,ω)\Gamma(\mathbf{r},\mathbf{r}',\omega) [and, consequently, also P(r,r′)P(\mathbf{r},\mathbf{r}')] is independent of the longitudinal electron wavefunction (along the e-beam direction) [30]. Therefore, for simplicity, we consider a point-like electron (extremely compressed longitudinal wavefunction), which can be treated as a classical current j(r,ω)=−ez^eiωz/vδ(R−R′)\mathbf{j}(\mathbf{r},\omega)=-e\hat{\mathbf{z}}e^{i\omega z/v}\delta(\mathbf{R}-\mathbf{R}'), where R′\mathbf{R}' denotes the transverse electron coordinates. In addition, we evaluate P(r,r′,ω)=P(R,z,R′,z′,ω)P(\mathbf{r},\mathbf{r}',\omega)=P(\mathbf{R},z,\mathbf{R}',z',\omega) at the same plane z′=zz'=z for both paths, so we drop the dependence on zz for clarity. The field induced at a position r\mathbf{r} by the current above is given by

E(r,R′,ω)=4πieω∫−∞∞dz′eiωz′/vG(r,r′,ω)⋅z^,\mathbf{E}(\mathbf{r},\mathbf{R}',\omega)=4\pi i e\omega\int_{-\infty}^{\infty}dz'e^{i\omega z'/v}G(\mathbf{r},\mathbf{r}',\omega)\cdot\hat{\mathbf{z}},

from which we obtain the decoherence probability by recasting Eq. (12) into

Γ(R,R′,ω)=12[Γ~(R,R′,ω)+Γ~(R′,R,ω)](13)\Gamma(\mathbf{R},\mathbf{R}',\omega)=\frac{1}{2}\left[\widetilde{\Gamma}(\mathbf{R},\mathbf{R}',\omega)+\widetilde{\Gamma}(\mathbf{R}',\mathbf{R},\omega)\right] \tag*{(13)}

with

Γ~(R,R′,ω)=eπℏω∫−∞∞dzRe⁡{e−iωz/vEz(r,R′,ω)}.(14)\widetilde{\Gamma}(\mathbf{R},\mathbf{R}',\omega)=\frac{e}{\pi\hbar\omega}\int_{-\infty}^{\infty}dz\operatorname{Re}\left\{e^{-i\omega z/v}E_z(\mathbf{r},\mathbf{R}',\omega)\right\}. \tag*{(14)}

This result provides a clearer physical interpretation of Γ(R,R′,ω)\Gamma(\mathbf{R},\mathbf{R}',\omega), which represents the probability that two electrons at lateral positions R\mathbf{R} and R′\mathbf{R}' exchange quanta of energy ℏω\hbar\omega. For clarity, we stress that we are considering single electrons throughout this work (i.e., low e-beam currents, as employed in electron microscopes), so Γ(R,R′,ω)\Gamma(\mathbf{R},\mathbf{R}',\omega) refers to different lateral positions of the same electron. We note that Eqs. (11) are suitable for numerical implementation (e.g., using boundary-element methods [31]), enabling decoherence modeling in geometries where analytical approaches are unavailable.

The EELS and decoherence probabilities can exceed unity, but this does not represent a physical problem. In particular, the decoherence probability enters through an exponential in Eq. (1), and thus, large values exceeding unity must be understood as a substantial depletion of coherence. Likewise, the EELS probability must be understood as an average number of excitations produced per electron. For example, considering a single bosonic mode initially prepared in the ground state, the post-interaction populations of number states follow a Poissonian distribution whose average is given by the frequency-integrated EELS probability [32, 33, 34].

For our two-path electron (see Figure 1b), decoherence becomes significant when the excitations generated by path AA (lateral position RA\mathbf{R}_A) are sufficiently different from those produced by path BB (at RB\mathbf{R}_B). The decoherence probability Eq. (9a) grows with the local EELS probabilities at RA\mathbf{R}_A and RB\mathbf{R}_B (R=R′\mathbf{R}=\mathbf{R}' terms), while the cross-term (R≠R′\mathbf{R}\ne\mathbf{R}') mitigates decoherence for modes excited with similar amplitude and phase at both positions. Strongly localized modes, specifically those with wavelengths or propagation lengths much shorter than ∣R−R′∣|\mathbf{R}-\mathbf{R}'|, are therefore expected to dominate the decoherence process. This motivates a spectral decomposition of the decoherence probability. Consequently, we define the frequency-resolved decoherence probability P(R,R′,ω)P(\mathbf{R},\mathbf{R}',\omega) as the integrand in Eq. (9a), such that P(R,R)=∫0∞dω P(R,R′,ω)P(\mathbf{R},\mathbf{R})=\int_{0}^{\infty}d\omega\,P(\mathbf{R},\mathbf{R}',\omega).

Besides the Bose-Einstein factor in Eq. (9a), there is a temperature dependence in the dielectric properties of the materials that form the specimen through the combined effect of lattice expansion, reduced excitation lifetimes, and changes in the electronic band structures. Although we do not enter into the details of these additional effects, we comment briefly on their relative importance. In metals, thermal lattice expansion and changes in the effective electron mass induce a small shift in the plasma frequency, while enhanced electron–phonon scattering significantly increases the damping rate [35, 36]. In dielectrics such as LiF, where lattice vibrations dominate the response, thermal expansion weakens the effective lattice force constants, causing a redshift in the resonance frequencies. Simultaneously, temperature-enhanced anharmonic phonon–phonon scattering increases optical damping [37, 38]. Finally, in transition metal dichalcogenides (e.g., MoS2_2), excitons dominate the visible optical response, and temperature changes produce substantial redshifts and broadenings of the resonance peaks via bandgap renormalization and enhanced exciton–phonon scattering [39, 40, 41]. Additionally, thermal energy can accelerate exciton dissociation into free carriers, resulting in a loss of spectral weight [39]. All of these thermal effects can be readily incorporated into our formalism by employing temperature-dependent permittivities.

Electron decoherence in bulk media

We study the decoherence experienced by an electron propagating inside an infinite homogeneous medium characterized by a momentum- and frequency-dependent dielectric tensor ϵ(q,ω)\boldsymbol{\epsilon}(\mathbf{q},\omega). We separate the wavevector q=Q+qzz^\mathbf{q}=\mathbf{Q}+q_z\hat{\mathbf{z}} into transverse and longitudinal components Q=(qx,qy)\mathbf{Q}=(q_x,q_y) and qzq_z relative to the propagation direction zz. Assuming translational invariance, the dielectric tensor can be written

or can be written

ϵ(q,ω)=ϵlon(q,ω)q^⊗q^+ϵtr(q,ω)(eQs⊗eQs+eQp⊗eQp)(15)\boldsymbol{\epsilon}(\mathbf{q},\omega)=\epsilon_{\mathrm{lon}}(\mathbf{q},\omega)\hat{\mathbf{q}}\otimes\hat{\mathbf{q}}+\epsilon_{\mathrm{tr}}(\mathbf{q},\omega)\left(\mathbf{e}_{Q_s}\otimes\mathbf{e}_{Q_s}+\mathbf{e}_{Q_p}\otimes\mathbf{e}_{Q_p}\right) \tag*{(15)}

in terms of longitudinal (lon) and transverse (tr) components. Here, q^=q/q\hat{\mathbf{q}}=\mathbf{q}/q, eQs=z^×Q^=(−qyx^+qxy^)/Q\mathbf{e}_{Q_s}=\hat{\mathbf{z}}\times\hat{\mathbf{Q}}=(-q_y\hat{\mathbf{x}}+q_x\hat{\mathbf{y}})/Q, eQp=eQs×q^=(qzQ^−Qz^)/q\mathbf{e}_{Q_p}=\mathbf{e}_{Q_s}\times\hat{\mathbf{q}}=(q_z\hat{\mathbf{Q}}-Q\hat{\mathbf{z}})/q, and Q^=Q/Q\hat{\mathbf{Q}}=\mathbf{Q}/Q. The corresponding real-space representation is ϵ(r,r′,ω)=(2π)−3∫d3q eiq⋅(r−r′)ϵ(q,ω)\boldsymbol{\epsilon}(\mathbf{r},\mathbf{r}',\omega)=(2\pi)^{-3}\int d^3q\,e^{i\mathbf{q}\cdot(\mathbf{r}-\mathbf{r}')}\boldsymbol{\epsilon}(\mathbf{q},\omega) [see Eq. (8)].

We evaluate the decoherence probability by applying Eqs. (9a) and (12) to an electron prepared in a coherent superposition of two straight-line paths separated by a distance d⊥d_\perp (Figure 2a). To this end, we first obtain the Green tensor G=Gbulk\mathbf{G}=\mathbf{G}^{\mathrm{bulk}} by rewriting Eq. (8) in momentum space as q×q×Gbulk(q,ω)+k2ϵ(q,ω)Gbulk(q,ω)=1/c2\mathbf{q}\times\mathbf{q}\times\mathbf{G}^{\mathrm{bulk}}(\mathbf{q},\omega)+k^2\boldsymbol{\epsilon}(\mathbf{q},\omega)\mathbf{G}^{\mathrm{bulk}}(\mathbf{q},\omega)=1/c^2. For the dielectric tensor in Eq. (15), we find the solution

Figure 2. Bulk contribution to electron decoherence. (a) An electron prepared in a two-path superposition propagates along a distance LL inside the bulk of a material characterized by a nonlocal dielectric tensor ϵ(q,ω)\epsilon(q,\omega). The two paths are separated by a distance d⊥d_{\perp}. (b) Frequency-resolved decoherence probability P(ω)P(\omega) for a 200 keV electron under the configuration of panel (a) when the material is Au, Al, or LiF at room temperature (T=300T = 300 K). We consider d⊥=1 μmd_{\perp} = 1\ \mu\mathrm{m} (solid curves) and d⊥→∞d_{\perp} \to\infty (dotted curves, coinciding with the EELS probability for a single-path electron). (c) Comparison between the decoherence probability obtained for Al with inclusion of retardation [taken from (b)] and the nonretarded calculation (i.e., taking c→∞c \to\infty). (d) Temperature dependence of the frequency-integrated decoherence probability PP for d⊥=0.3 μmd_{\perp} = 0.3\ \mu\mathrm{m} (solid curves) and d⊥=30 μmd_{\perp} = 30\ \mu\mathrm{m} (dashed curves). The half-collection angle is set to φout=10\varphi_{\mathrm{out}} = 10 mrad throughout this work.

Gbulk(q,ω)=q^⊗q^ω2ϵlon(q,ω)+eQs⊗eQs+eQp⊗eQp[ω2ϵtr(q,ω)−c2q2]\mathbf{G}^{\mathrm{bulk}}(\mathbf{q},\omega)=\frac{\hat{\mathbf{q}}\otimes\hat{\mathbf{q}}}{\omega^2\epsilon_{\mathrm{lon}}(\mathbf{q},\omega)}+\frac{\mathbf{e}_{Q_s}\otimes\mathbf{e}_{Q_s}+\mathbf{e}_{Q_p}\otimes\mathbf{e}_{Q_p}}{\left[\omega^2\epsilon_{\mathrm{tr}}(\mathbf{q},\omega)-c^2q^2\right]}

We now proceed by moving back to real space via Gbulk(r,r′,ω)=(2π)−3∫d3q eiq⋅(r−r′)Gbulk(q,ω)\mathbf{G}^{\mathrm{bulk}}(\mathbf{r},\mathbf{r}',\omega)=(2\pi)^{-3}\int d^3q\,e^{i\mathbf{q}\cdot(\mathbf{r}-\mathbf{r}')}\mathbf{G}^{\mathrm{bulk}}(\mathbf{q},\omega), retaining the zzzz component, inserting it into Eq. (12) to obtain the nonlocal loss-probability, and substituting the result into Eq. (9a). We then obtain the bulk decoherence probability

Pbulk(d⊥)=∫0∞dω [2nT(ω)+1][Γbulk(0,ω)−Γbulk(d⊥,ω)](16)P^{\mathrm{bulk}}(d_\perp)=\int_0^\infty d\omega\,[2n_T(\omega)+1]\left[\Gamma^{\mathrm{bulk}}(0,\omega)-\Gamma^{\mathrm{bulk}}(d_\perp,\omega)\right] \tag*{(16)}

where

Γbulk(R,ω)=2e2Lπℏv2∫0∞Q dQJ0(QR)Im⁡{[1ϵtr(q,ω)−1ϵlon(q,ω)]1q2+[v2c2−1ϵtr(q,ω)]1q2−k2ϵtr(q,ω)}(17)\Gamma^{\mathrm{bulk}}(R,\omega)=\frac{2e^2L}{\pi\hbar v^2}\int_0^\infty Q\,dQJ_0(QR)\operatorname{Im}\left\{\left[\frac{1}{\epsilon_{\mathrm{tr}}(\mathbf{q},\omega)}-\frac{1}{\epsilon_{\mathrm{lon}}(\mathbf{q},\omega)}\right]\frac{1}{q^2}+\left[\frac{v^2}{c^2}-\frac{1}{\epsilon_{\mathrm{tr}}(\mathbf{q},\omega)}\right]\frac{1}{q^2-k^2\epsilon_{\mathrm{tr}}(\mathbf{q},\omega)}\right\} \tag*{(17)}

is the nonlocal energy-loss probability for an electron moving in a bulk medium and two positions separated by a distance RR [29]. To study a representative set of materials, we apply Eqs. (16) and (17) to a dielectric (LiF), as well as simple (Al) and noble (Au) metals, where we explore the contributions of phononic and plasmonic modes, respectively. For LiF, we adopt the local approximation and assume isotropic permittivity components ϵlon(ω)=ϵtr(ω)\epsilon_{\mathrm{lon}}(\omega)=\epsilon_{\mathrm{tr}}(\omega). We rely on experimental data from Ref. [42], which we extend to lower frequencies by fitting the optical response to a two-oscillator model as

ϵLiF(ω)=ϵ∞+∑i=12siωi2ωi2−ω(ω+iηi)(18)\epsilon^{\mathrm{LiF}}(\omega)=\epsilon_\infty+\sum_{i=1}^{2}\frac{s_i\omega_i^2}{\omega_i^2-\omega(\omega+i\eta_i)} \tag*{(18)}

with parameters ϵ∞=1.96\epsilon_\infty=1.96, ℏω1=38\hbar\omega_1=38 meV, ℏω2=62\hbar\omega_2=62 meV, ℏη1=2.16\hbar\eta_1=2.16 meV, ℏη2=10.7\hbar\eta_2=10.7 meV, s1=6.67s_1=6.67, and s2=0.116s_2=0.116 [42]. We specifically apply Eq. (18) to frequencies below the lowest-energy phonon peak at ℏω=38\hbar\omega=38 meV, noting that it reproduces the experimental data accurately near the phonon resonances, although it does not capture higher-frequency features, which become increasingly relevant beyond LiF’s band gap of 13.6 eV.

For metals, nonlocal effects may become relevant due to high-momentum transfers associated with the strong lateral focusing of the electron. To capture these effects, we adopt an independent-electron picture for the conduction electrons, whose longitudinal response is modelled in the random-phase approximation (RPA) [43], corrected to conserve the local electron density as prescribed by Mermin [44]. We thus set ϵlon(q,ω)\epsilon_{\mathrm{lon}}(q,\omega) to the Mermin dielectric function ϵM(q,ω)\epsilon^{\mathrm{M}}(q,\omega) (see Appendix B). The transverse component, which accounts for polarization associated with longer-wavelength modes, is approximated by a local Drude model

ϵD(ω)=1−ωp2ω(ω+iη),(19)\epsilon^{\mathrm{D}}(\omega) = 1 - \frac{\omega_{\mathrm{p}}^2}{\omega(\omega+ i\eta)}, \tag*{(19)}

where ωp\omega_{\mathrm{p}} is the classical plasma frequency and η\eta is the damping rate. We set ℏωp=15\hbar\omega_{\mathrm{p}} = 15 eV and ℏη=0.6\hbar\eta= 0.6 eV for Al [42], and ℏωp=9\hbar\omega_{\mathrm{p}} = 9 eV and ℏη=0.05\hbar\eta= 0.05 eV for Au [45]. Note that the Mermin function rigorously reduces to the Drude form in the local limit [i.e., ϵM(q→0,ω)=ϵD(ω)].\epsilon^{\mathrm{M}}(q \to0,\omega) = \epsilon^{\mathrm{D}}(\omega)]. This model produces excellent results for Al, where the response of inner bands plays a negligible role.

In contrast to Al, valence-band screening becomes important in Au, particularly in the plasmonic region. To address this issue, we adopt a model that treats the valence-band response in the local approximation and conduction electrons in the Lindhard+Mermin model [46]. Specifically, we estimate the contribution of valence bands to the permittivity from the difference between the experimentally measured frequency-dependent dielectric function ϵexp(ω)\epsilon^{\mathrm{exp}}(\omega) [45] and its fit to the Drude model. In addition, we retain the Mermin model to describe conduction electrons. The resulting longitudinal dielectric function takes the form [46]

ϵlonAu(q,ω)=ϵexp(ω)−ϵD(ω)+ϵM(q,ω).(20)\epsilon_{\mathrm{lon}}^{\mathrm{Au}}(q,\omega) = \epsilon^{\mathrm{exp}}(\omega) - \epsilon^{\mathrm{D}}(\omega) + \epsilon^{\mathrm{M}}(q,\omega). \tag*{(20)}

In addition, the transverse permittivity of Au is identified as ϵtrAu(q,ω)=ϵexp(ω)\epsilon_{\mathrm{tr}}^{\mathrm{Au}}(q,\omega) = \epsilon^{\mathrm{exp}}(\omega).

In the local approximation, the integral over QQ can be performed analytically for a finite path separation ($R > 0), yielding the result

Γbulk(R,ω)=2e2Lπℏv2Im⁡{(v2c2−1ϵ)K0(ωRvγϵ)},(21)\Gamma^{\mathrm{bulk}}(R,\omega) = \frac{2e^2L}{\pi\hbar v^2}\operatorname{Im}\left\{\left(\frac{v^2}{c^2} - \frac{1}{\epsilon}\right)K_0\left(\frac{\omega R}{v\gamma_{\epsilon}}\right)\right\}, \tag*{(21)}

where γϵ=1/1−ϵ(ω)v2/c2\gamma_{\epsilon} = 1/\sqrt{1-\epsilon(\omega)v^2/c^2}, ϵ\epsilon is understood to have an implicit dependence on ω\omega, K0K_0 is a Bessel function of the second kind, and we have used the identity ∫0∞dQ QJ0(QR)/(Q2+a2)=K0(aR)\int_0^\infty dQ\,QJ_0(QR)/(Q^2+a^2) = K_0(aR) (see Eq. 6.532-(4) in Ref. [47]). This expression diverges logarithmically in the R→0R \to0 limit due to the αQ−2\alpha Q^{-2} scaling of the integrand for large wavevector QQ, which can take arbitrarily large values for a point-like electron. In practice, momentum exchanges must remain finite, and we introduce a cutoff Qc≈φoutq0Q_{\mathrm{c}} \approx\varphi_{\mathrm{out}}q_0 defined by the half-collection angle of the electron microscope [29], which we set to 10 mrad from here on. By integrating up to QcQ_{\mathrm{c}} for R=0R = 0, we find

Γbulk(0,ω)=e2Lπℏv2Im⁡{(v2c2−1ϵ)log⁡[1+(vγϵQcω)2]}.(22)\Gamma^{\mathrm{bulk}}(0,\omega) = \frac{e^2L}{\pi\hbar v^2}\operatorname{Im}\left\{\left(\frac{v^2}{c^2} - \frac{1}{\epsilon}\right)\log\left[1+\left(\frac{v\gamma_{\epsilon}Q_{\mathrm{c}}}{\omega}\right)^2\right]\right\}. \tag*{(22)}

Throughout this paper, we introduce this cutoff to obtain results within the local approximation.

For an infinite path separation (d⊥→∞d_{\perp} \to\infty), the decoherence probability exhibits a ω−1\omega^{-1} divergence for all three materials, as shown in Figure 2b, where we plot the spectral dependence of Pbulk(d⊥,ω)P^{\mathrm{bulk}}(d_{\perp},\omega) for 200 keV electrons. This can be understood by examining the dielectric functions for both types of materials in the low frequency limit [Eq. (15) for LiF and Eq. (19) for the metals]. For LiF, the static permittivity ϵLiF(0)=ϵ∞+s1+s2\epsilon^{\mathrm{LiF}}(0) = \epsilon_{\infty} + s_1 + s_2 is purely real, and therefore, Γbulk\Gamma^{\mathrm{bulk}} vanishes unless we have Cherenkov losses under the condition v/c>1/ϵLiF(0)≈0.33v/c > 1/\sqrt{\epsilon^{\mathrm{LiF}}(0)} \approx0.33, which makes γϵ\gamma_{\epsilon} imaginary. Because we consider 200 keV electrons (v/c≈0.7v/c \approx0.7) throughout this work, the Cherenkov condition at zero frequency is always satisfied. In contrast, for metals, the permittivity has a divergent imaginary part at low frequencies, and thus, both K0K_0 in Eq. (21) and the logarithm in Eq. (22) acquire nonzero imaginary parts, regardless of the electron velocity; a divergence is thus observed at low ω\omega when multiplying the zero-temperature EELS probability by nT(ω→0)∝ω−1n_T(\omega\to0) \propto\omega^{-1}. Interestingly, this divergence is a retardation effect, as it vanishes in the c→∞c \to\infty limit, as shown in Figure 2c, where the low-frequency behavior is dramatically different with and without inclusion of retardation. Although we only compare retarded and nonretarded limits for Al here, Au follows a similar trend. Conversely, for LiF, the nonretarded limit makes the Cherenkov condition impossible to be satisfied, thereby suppressing the divergence.

For finite path separations, the divergence is regularized by the cross term in Eq. (9a), which suppresses contributions from low-energy excitations when their wavelengths are large compared with d⊥d_{\perp}, so that the electron–mode interaction produces similar amplitudes in the two paths (i.e., they do not provide which-way information). This behavior is clearly observed in Figure 2b. Incidentally, P(ω→0,d⊥)P(\omega\to0,d_{\perp}) is much larger for metals than for LiF, as expected from the fact that Im⁡{ϵ}\operatorname{Im}\{\epsilon\} diverges as ω−1\omega^{-1} in the former, while it vanishes in the latter.

The ω−1\omega^{-1} divergence in the limit of infinite path separation is also a finite-temperature effect, as it originates from thermal occupation of modes with energies below kBTk_{\mathrm{B}}T, and indeed, the divergence disappears in the T=0T=0 limit (see Supplementary Figure S6). Consequently, the temperature dependence primarily enters through the low-frequency part of the spectrum.

Frequency-resolved decoherence probabilities for bulk contributions

Figure S6. Zero-temperature limit: Bulk contribution to electron decoherence (a) Frequency-resolved decoherence probability P(ω)P(\omega) for a 200 keV electron under the configuration of Figure 2a in the main text when the material is Au, Al, or LiF at T=0T = 0. We consider d⊥=1 μmd_\perp= 1\,\mu\mathrm{m} (solid curves) and d⊥→∞d_\perp\to\infty (dotted curves, coinciding with the EELS probability for a single-path electron). (b) Comparison between the decoherence probability obtained for Al under the configuration in Figure 2a of the main text at T=0T = 0 with inclusion of retardation [taken from (a)] and the nonretarded calculation (i.e., taking c→∞c \to\infty).

In addition, P(ω)P(\omega) displays different material-dependent features: for Al, the only discernible peak is the bulk plasmon at ∼15\sim15 eV, while Au displays a broader response for energies starting at its bulk plasmon of ∼2.5\sim2.5 eV; for LiF, we find two peaks corresponding to optical phonons at ∼50\sim50 meV and interband transitions above the electronic band gap of 13.6 eV.

Surface contribution to electron decoherence

The linearity of Eq. (8) allows us to decompose the Green tensor as G=Gbulk+GsurfG=G^{\mathrm{bulk}}+G^{\mathrm{surf}}, where GbulkG^{\mathrm{bulk}} is defined in each homogeneous region of the specimen as the solution in an infinite medium (e.g., in vacuum or inside a material), while GsurfG^{\mathrm{surf}} compensates for the presence of surfaces and interfaces. Consequently, since Eq. (12) for the nonlocal EELS probability is linear in the Green tensor, we can write it as the sum of bulk and surface contribution Γ=Γbulk+Γsurf\Gamma=\Gamma^{\mathrm{bulk}}+\Gamma^{\mathrm{surf}}, and similarly, from Eq. (9a) for the decoherence probability, P=Pbulk+PsurfP=P^{\mathrm{bulk}}+P^{\mathrm{surf}}. While in § 1.2 we have discussed bulk effects for electrons propagating fully inside the bulk of a material, in the present section we study surface effects under two representative configurations for two-path electrons, depending on whether the e-beam is oriented parallel or perpendicular to the surface. In what follows, we include the bulk contribution using the formalism introduced in § 1.2, while surface contribution are calculated within the local approximation, using Eq. (18) for the permittivity of LiF, the Drude approximation in Eq. (19) for Al, and measured data of ϵexp⁡(ω)\epsilon_{\exp}(\omega) [45] for Au.

whether the e-beam is oriented parallel or perpendicular to the surface. In what follows, we include the bulk contribution using the formalism introduced in § 1.2, while surface contribution are calculated within the local approximation, using Eq. (18) for the permittivity of LiF, the Drude approximation in Eq. (19) for Al, and measured data of ϵexp⁡(ω)\epsilon_{\exp}(\omega) [45] for Au.

Surface-parallel electron beams

We consider the configuration sketched in Figure 3a, where a two-path electron propagates a distance LL along xx (instead of zz, which is reserved for the surface normal below), parallel to the surface of a semi-infinite material (z=0z = 0 plane), with one path moving inside the medium (z<0z < 0 region) and the other one in vacuum. The electron can couple to surface excitations and also Cherenkov radiation inside the material [48].

Figure 3. Surface contribution to electron decoherence: Parallel e-beam configuration. (a) An electron prepared in a two-path superposition propagates along a distance LL parallel to the planar surface of a semi-infinite medium of local permittivity ϵ(ω)\epsilon(\omega). One of the electron paths is inside the material, and the other is in the surrounding vacuum. The path–surface separations are d1d_1 and d0d_0, respectively. (b) Surface contribution to the frequency-resolved decoherence probability Psurf(ω)P^{\mathrm{surf}}(\omega) for a 200 keV electron under the configuration of panel (a) when the material is Au, Al, or LiF at room temperature (T=300T=300 K). We set d0=d1=0.5d_0=d_1=0.5 μ\mum. As PsurfP^{\mathrm{surf}} becomes negative for some frequencies, we plot both PsurfP^{\mathrm{surf}} (solid curves) and −Psurf-P^{\mathrm{surf}} (broken curves). The broken black line indicates a ω−1\omega^{-1} slope, matching that of the infrared divergence, which is regularized when the bulk contributions are added. (c) Decoherence probability PP (frequency-integrated bulk+surface) for LiF at T=300T=300 K as a function of d0d_0 and d1d_1. (d) Partial decoherence probability PcutoffP_{\mathrm{cutoff}} obtained by limiting the spectral integral to ℏω≤1\hbar\omega\leq1 eV under the same conditions as in panel (c).

Bulk components are directly included in the inner path following § 1.2, while surface contribution require some analysis of the corresponding electromagnetic Green tensor, which can be obtained by solving Eq. (8) and imposing the the continuity of x−yx-y components of G(r,r′,ω)\mathbf{G}(\mathbf{r},\mathbf{r}',\omega) and ∇×G(r,r′,ω)\nabla\times\mathbf{G}(\mathbf{r},\mathbf{r}',\omega) at z=0z = 0 [49]. Alternatively, by noticing that this quantity coincides with the electric field produced at r\mathbf{r} by a dipole of strength −1/(4πω2)-1/(4\pi\omega^2) placed at r′\mathbf{r}' and oscillating with frequency ω\omega, we can also obtain the Green tensor from the dipole-induced electric field projected in in-plane momentum components. Following this procedure, we obtain the xxxx component (i.e., along the e-beam direction) of the surface contribution as

Gxxsurf(r,r′,ω)=−i8π2ω2∫d2QQ2eiQ⋅(R−R′)g(Q,z,z′,ω),(23)G_{xx}^{\mathrm{surf}}(\mathbf{r},\mathbf{r}',\omega) = \frac{-i}{8\pi^{2}\omega^{2}} \int\frac{\mathrm{d}^{2}\mathbf{Q}}{Q^{2}} e^{i\mathbf{Q}\cdot(\mathbf{R}-\mathbf{R}')}g(\mathbf{Q},z,z',\omega), \tag*{(23)}

where

g(Q,z,z′,ω)={eiqz0(z+z′)(rs01qy2k2−qx2qz02rp01)/qz0,z,z′>0,ei(qz0z−qz1z′)(ts10qy2k2+Q2qz0qz1tp10/ϵ)/qz1,z>0,z′<0,e−iqz1(z+z′)(rs10qy2k2−qx2qz12rp10/ϵ)/qz1,z,z′<0.(24)g(\mathbf{Q},z,z',\omega)= \begin{cases} e^{iq_{z0}(z+z')}\left(r_{s}^{01}q_{y}^{2}k^{2}-q_{x}^{2}q_{z0}^{2}r_{p}^{01}\right)/q_{z0}, & z,z'>0,\\ e^{i(q_{z0}z-q_{z1}z')}\left(t_{s}^{10}q_{y}^{2}k^{2}+Q^{2}q_{z0}q_{z1}t_{p}^{10}/\sqrt{\epsilon}\right)/q_{z1}, & z>0,z'<0,\\ e^{-iq_{z1}(z+z')}\left(r_{s}^{10}q_{y}^{2}k^{2}-q_{x}^{2}q_{z1}^{2}r_{p}^{10}/\epsilon\right)/q_{z1}, & z,z'<0. \end{cases} \tag*{(24)}

Here, qzj=k2ϵj−Q2q_{zj}=\sqrt{k^{2}\epsilon_{j}-Q^{2}} (with the square root yielding a positive imaginary part) are the out-of-plane wavevector components in vacuum (j=0j=0, ϵ0=1\epsilon_{0}=1) and inside the material (j=1j=1, ϵ1≡ϵ\epsilon_{1}\equiv\epsilon). Also, we use Fresnel’s reflection and transmission coefficients

rsjj′=qzj−qzj′qzj+qzj′,rpjj′=ϵj′qzj−ϵjqzj′ϵj′qzj+ϵjqzj′,r_{s}^{jj'}=\frac{q_{zj}-q_{zj'}}{q_{zj}+q_{zj'}}, \qquad r_{p}^{jj'}=\frac{\epsilon_{j'}q_{zj}-\epsilon_{j}q_{zj'}}{\epsilon_{j'}q_{zj}+\epsilon_{j}q_{zj'}},
tsjj′=2qzjqzj+qzj′,tpjj′=2ϵjϵj′qzjϵj′qzj+ϵjqzj′,t_{s}^{jj'}=\frac{2q_{zj}}{q_{zj}+q_{zj'}}, \qquad t_{p}^{jj'}=\frac{2\sqrt{\epsilon_{j}\epsilon_{j'}}q_{zj}}{\epsilon_{j'}q_{zj}+\epsilon_{j}q_{zj'}},

for s and p polarization and incidence from medium jj on the j/j′j/j' interface. For z<0z<0 and z′>0z'>0, reciprocity allows us to write Gxxsurf(r,r′,ω)=Gxxsurf(r′,r,ω)G_{xx}^{\mathrm{surf}}(\mathbf{r},\mathbf{r}',\omega)=G_{xx}^{\mathrm{surf}}(\mathbf{r}',\mathbf{r},\omega).

Inserting Eq. (23) into Eq. (12), we obtained the nonlocal EELS probability

Γsurf(z,z′,ω)=2Le2πℏω2∫0∞dqyQ2Re⁡{g(ω/v,qy,z,z′,ω)},\Gamma^{\mathrm{surf}}(z,z',\omega)=\frac{2Le^{2}}{\pi\hbar\omega^{2}}\int_{0}^{\infty}\frac{\mathrm{d}q_{y}}{Q^{2}}\operatorname{Re}\{g(\omega/v,q_{y},z,z',\omega)\},

where the condition qx=ω/vq_{x}=\omega/v in gg [Eq. (24)] is imposed by phase-matching in the electron–surface interaction for a parallel trajectory, producing an overall factor LL for the interaction length. Additionally, bulk processes only contribute to the inner path through Γbulk(z,z,ω)\Gamma^{\mathrm{bulk}}(z,z,\omega) with z<0z<0. Finally, the decoherence probability is given by inserting these expressions into Eq. (9a).

In Figure 3b, we show the frequency-resolved decoherence probability contributed by surface excitations, which can take negative values, although the overall probability including bulk processes must remain non-negative. This effect is particularly relevant at low frequencies, where the surface contribution cancels the divergence of the bulk term Pbulk(ω)P^{\mathrm{bulk}}(\omega) for finite path separations. In the infinite separation limit (∣z−z′∣=d⊥→∞|z-z'|=d_{\perp}\rightarrow\infty), the cross term vanishes in Eq. (9a), and the spectrally resolved decoherence probability retains an infrared divergence as ω−1\omega^{-1} for T≠0T\ne0 (see Supplementary Fig. S8 for results in the zero-temperature limit). Aside from this low-frequency behavior, the results are qualitatively similar for a configuration in which both paths are inside the material, but with a notable offset in the overall value of PP.

A high decoherence probability under the configuration of Figure 3 indicates a higher likelihood of interaction with bulk modes for the path inside the material compared to the one in vacuum. An interesting behavior is observed when comparing excitations from different parts of the spectrum by discarding transmitted electrons that have lost energy above a certain cutoff. In LiF, without any energy cutoff (Figure 3c), PP remains nearly constant over a wide range of d0d_{0} and d1d_{1} values because it is dominated by electronic excitations above the band gap (i.e., ℏω>13.6 eV\hbar\omega>13.6\ \mathrm{eV}, see Figure 2b), which are strongly localized (i.e., their associated free-space wavelength is λ<90 nm\lambda<90\ \mathrm{nm}). In contrast, the decoherence probability PcutoffP_{\mathrm{cutoff}} for a cutoff ℏω=1 eV\hbar\omega=1\ \mathrm{eV} (Figure 3d) exhibits a strong spatial dependence over distances in the micrometer range, as expected from the dominant contributions below ∼100 meV\sim100\ \mathrm{meV}, or equivalently λ>12 μm\lambda>12\ \mu\mathrm{m}. With and without a cutoff in PP, decoherence is maximum when the external path lies near the surface, and the inner one lies deep inside the material, such that the former couples predominantly to surface modes that decay quickly inside the material, while the latter interacts primarily with bulk modes that do not reach the surface.

Thin films traversed under normal incidence

We consider a scenario of particular relevance to holography experiments in transmission electron microscopy, where a two-path electron with transverse separation d⊥d_{\perp} traverses a planar film of thickness LL under normal incidence conditions (Figure 4a). The electron interacts with both surface and guided modes supported by the film. In addition, because surface crossing breaks momentum conservation, it also couples to radiative modes, generating transition radiation [50].

Figure 4. Surface contribution to electron decoherence: Perpendicular e-beam configuration. (a) An electron prepared in a superposition of two paths separated by a distance d⊥d_\perp traverses a homogeneous film (thickness LL) made of a material of local permittivity ϵ(ω)\epsilon(\omega). We consider normal e-beam incidence relative to the film. (b) Frequency-resolved decoherence probability at T=300T=300 K for a 200 keV electron under the configuration of panel (a) when the material is Au, Al, or LiF. We set d⊥=1d_\perp=1 μ\mum (solid curves) and d⊥→∞d_\perp\to\infty (broken curves, coinciding with the EELS probability). (c) Comparison of retarded and nonretarded regimes for the frequency-resolved decoherence probability in an Al film. (d) Full surface contribution to the frequency-resolved decoherence probability Psurf(ω)P^{\mathrm{surf}}(\omega) (solid curves) compared with the partial contribution to this quantity arising from guided modes (dashed curves). (e) Ratio Psurf/PbulkP^{\mathrm{surf}}/P^{\mathrm{bulk}} of the frequency-integrated surface to bulk decoherence probabilities as a function of d⊥d_\perp for different values of the temperature. (f) Temperature dependence of the decoherence probability for d⊥=10d_\perp=10 μ\mum.

In this geometry, the explicit expression for the Green tensor becomes cumbersome, so instead we compute the decoherence probability using Eq. (11b), based on the electron-induced electric field. Given the two-dimensional symmetry of the system, the scattering problem can be solved independently for each in-plane wavevector component Q\mathbf{Q}, so we expand the field produced by an electron following a trajectory with transverse coordinates R′\mathbf{R}' as E(r,ω)=(2π)−2∫d2QeiQ⋅(R−R′)E(Q,z,ω)\mathbf{E}(\mathbf{r},\omega)=(2\pi)^{-2}\int d^{2}Q e^{i\mathbf{Q}\cdot(\mathbf{R}-\mathbf{R}')} \mathbf{E}(\mathbf{Q},z,\omega) in terms of such components. Following Ref. [48], we write a solution of the form E=Ebulk+Esurf\mathbf{E}=\mathbf{E}^{\mathrm{bulk}}+\mathbf{E}^{\mathrm{surf}}, where

Ejbulk(Q,z,ω)=4πievϵjeiωz/vq−kϵjv/cq2−k2ϵj(25)\mathbf{E}_{j}^{\mathrm{bulk}}(\mathbf{Q},z,\omega)=\frac{4\pi i e}{v\epsilon_{j}}e^{i\omega z/v}\frac{\mathbf{q}-k\epsilon_{j}\mathbf{v}/c}{q^{2}-k^{2}\epsilon_{j}} \tag*{(25)}

with qz=ω/vq_{z}=\omega/v is the bulk field generated by an electron moving in an infinite medium jj (=0=0 and 11 for vacuum and the material), while

Esurf(Q,z,ω)={Ae−iqz0zeQp0−,z<0,∑±C±e±iqz1zeQp1±,0<z<L,Beiqz0zeQp0+,z>L,(26)\mathbf{E}^{\mathrm{surf}}(\mathbf{Q},z,\omega)= \begin{cases} A e^{-iq_{z0}z}\mathbf{e}_{Qp}^{0-}, & z<0,\\ \sum_{\pm}C_{\pm}e^{\pm iq_{z1}z}\mathbf{e}_{Qp}^{1\pm}, & 0<z<L,\\ B e^{iq_{z0}z}\mathbf{e}_{Qp}^{0+}, & z>L, \end{cases} \tag*{(26)}

is the surface-scattered field (with the film contained in the 0<z<L0<z<L region), which consists of outgoing waves in the vacuum regions and a superposition of counterpropagating waves inside the film. For the perpendicular trajectory under consideration, the field is entirely made of p-polarized components with polarization vectors defined as eQpj±=(±qzjQ−Q2z^)/(kQϵj)\mathbf{e}_{Qp}^{j\pm}=(\pm q_{zj}\mathbf{Q}-Q^{2}\hat{\mathbf{z}})/(kQ\sqrt{\epsilon_{j}}) with qzj=k2ϵj−Q2q_{zj}=\sqrt{k^{2}\epsilon_{j}-Q^{2}}. The amplitudes AA, BB, and C±C_{\pm} are determined from the continuity of the parallel components of the total electric and magnetic fields at z=0z=0 and z=Lz=L [48], as shown in Appendix C.

Inserting Eq. (26) into Eqs. (11a) and (11b), we find the following expression for the surface contribution to the nonlocal EELS probability:

Γsurf(R,ω)=ec2π2vℏω2∫0∞dQ Q2J0(QR)Im⁡{Aω/v+qz0−Be−i(ω/v−qz0)Lω/v−qz0+∑±C±[e−i(ω/v∓qz1)L−1]ϵ(ω/v∓qz1)}(27)\Gamma^{\mathrm{surf}}(R,\omega)=\frac{ec}{2\pi^{2}v\hbar\omega^{2}}\int_{0}^{\infty}dQ\,Q^{2}J_{0}(QR)\operatorname{Im}\left\{\frac{A}{\omega/v+q_{z0}}-\frac{B e^{-i(\omega/v-q_{z0})L}}{\omega/v-q_{z0}}+\sum_{\pm}\frac{C_{\pm}\left[e^{-i(\omega/v\mp q_{z1})L}-1\right]}{\sqrt{\epsilon}(\omega/v\mp q_{z1})}\right\} \tag*{(27)}

where the induced field has been integrated over zz. Finally, the decoherence probability obtained by inserting Γsurf(R,ω)\Gamma^{\mathrm{surf}}(R,\omega) into Eq. (9a). Incidentally, by using the bulk field given by Eq. (25) instead of Eq. (26), we recover Eq. (21) with LL identified as the film thickness, since vacuum regions do not contribute to Γbulk(R,ω)\Gamma^{\mathrm{bulk}}(R,\omega). Next, we present numerical results for PsurfP^{\mathrm{surf}} and P=Pbulk+PsurfP=P^{\mathrm{bulk}}+P^{\mathrm{surf}}.

Examining the frequency-resolved decoherence probability for a 200 keV electron traversing a 100 nm-thick film (Figure 4b), we again observe a low-ω\omega divergence in the limit of infinite path separation. For LiF, this divergence scales as ω−1\omega^{-1} at finite temperature and originates from the bulk contribution. In metals, the divergence is stronger, scaling as ω−1\omega^{-1} and ω−2\omega^{-2} at zero and finite temperature, respectively (see Supplementary Figure S8 for results in the zero-temperature limit). Introducing a finite path separation suppresses these divergences, rendering

Frequency-resolved decoherence probabilities for perpendicular e-beam configurations at zero temperature

Figure S8. Zero temperature limit: Surface contribution to electron decoherence: Perpendicular e-beam configuration. (a) Frequency-resolved decoherence probability P(ω)P(\omega) at T=0T = 0 for a 200 keV electron under the configuration of Figure 4a in the main text when the material is Au, Al, or LiF. We consider d⊥=1 μmd_\perp= 1\,\mu\mathrm{m} (solid curves) and d⊥→∞d_\perp\to\infty (dotted curves, coinciding with the EELS probability for a single-path electron). (b) Comparison between the decoherence probability at T=0T = 0 obtained for Al under the configuration in Figure 4a of the main text with inclusion of retardation [taken from (a)] and the nonretarded calculation (i.e., taking c→∞c \to\infty). (c) Full surface contribution to the frequency-resolved decoherence probability Psurf(ω)P^{\mathrm{surf}}(\omega) (solid curves) compared with the partial contribution to this quantity arising from guided modes (dashed curves) at T=0T = 0.

the decoherence probability integrable again. Notably, the logarithmic divergence associated with the bulk contribution in Al and Au (see Figure 2b) is canceled by the surface contribution. While we have fixed L=100L=100 nm throughout this section, we explore the dependence on LL in Supplementary Figure S9.

Figure S9. Effect of film thickness on electron decoherence in the perpendicular e-beam configuration. (a,d,g) Frequency-resolved decoherence probability at T=300T = 300 K for a 200 keV electron under the configuration of Figure 4a in the main text when the material is Au, Al, or LiF. We set d⊥=1d_{\perp} = 1 μ\mum (solid curves) and d⊥→∞d_{\perp} \to\infty (broken curves, coinciding with the EELS probability). (b,e,h) Ratio Psurf/PbulkP^{\mathrm{surf}}/P^{\mathrm{bulk}} of the frequency-integrated surface to bulk decoherence probabilities as a function of d⊥d_{\perp} for different values of the temperature. (c,f,i) Temperature dependence of the decoherence probability for d⊥=10d_{\perp} = 10 μ\mum. The film thickness is L=25L = 25 nm in (a-c), L=50L = 50 nm in (b-f), and L=75L = 75 nm in (g-i).

Retardation plays an important role in the observed divergences, as illustrated in Figure 4c, where we compare the decoherence probability for an Al film calculated with a finite speed of light to that obtained in the c→∞c\rightarrow\infty limit (with similar results observed for Au). Retardation enters through transition radiation [i.e., Q<kQ<k components of Eq. (25) for j=0j=0], which dominates the divergence behavior. In contrast, in the nonretarded limit, the divergence originates from the excitation of long-wavelength surface plasmons. Interestingly, although plasmons decay rapidly with distance, partial coherence is retained at finite path separations.

In LiF, we observe low-energy features (≲0.1\lesssim0.1 eV, see Figure 4b) associated with the excitation of phonon polaritons and dominated by guided modes [i.e., Q>kQ>k components in Eq. (26)], as we show in Figure 4d by comparing results for PsurfP^{\mathrm{surf}} performing the full integral over QQ with those obtained by only integrating above kk. In the ω→0\omega\to0 limit, PsurfP^{\mathrm{surf}} remains finite even at infinite path separation, showing that the divergence in PP for LiF (Figure 4b) originates exclusively from the bulk contribution.

For metals, the ω−2\omega^{-2} divergence becomes increasingly more pronounced at larger d⊥d_\perp, where wavelengths comparable to the path separation enhance decoherence. The comparison of surface and bulk contributions in Figure 4e shows that the latter remains dominant even for d⊥≈50d_\perp\approx50 μ\mum, but the relative weight of PsurfP^{\mathrm{surf}} increases with d⊥d_\perp, a trend that becomes more apparent at elevated

temperatures. Consequently, metals display an enhanced temperature dependence, with PP varying by >1%>1\% when moving from T=0T=0 to T=1000T=1000 K (Figure 4f). In contrast, for LiF, the surface contribution associated with phononic modes is reinforced at elevated temperatures, but does not produce a divergence as pronounced as in metals, while the high-energy bulk contribution still dominates (Figure 4e), leading to a similar fractional change of PP with temperature as in metals.

Temperature dependence

The strong temperature dependence of electron decoherence suggests its use for nanoscale thermometry. As shown above, the dependence is dramatic for low-frequency excitations (ℏω≲kBT\hbar\omega\lesssim k_{\mathrm{B}}T). However, when integrating over all frequencies, contributions from higher-energy modes often dominate, reducing the relative weight of the temperature dependence. The temperature sensitivity can be increased by suppressing these high-energy contributions through energy filtering of the e-beam after interaction, retaining only low-frequency excitations. Under these conditions, it is useful to introduce an energy-filtered decoherence probability

Materialω→0\omega\to0T=0T=0 Bulk / Interfaces ∥v\parallel\mathbf{v}T=0T=0 Interface ⊥v\perp\mathbf{v}T≠0T\ne0 Bulk / Interfaces ∥v\parallel\mathbf{v}T≠0T\ne0 Interface ⊥v\perp\mathbf{v}
Dielectrics (LiF)Γ(ω)∝\Gamma(\omega)\proptoω0\omega^{0}ω0\omega^{0}ω−1\omega^{-1}ω−1\omega^{-1}
P(ω)∝P(\omega)\proptoω\omegaω\omegaω0\omega^{0}ω0\omega^{0}
Metals (Au, Al)Γ(ω)∝\Gamma(\omega)\proptoω0\omega^{0}ω−1\omega^{-1}ω−1\omega^{-1}ω−2\omega^{-2}
P(ω)∝P(\omega)\proptoωlog⁡(ω)\omega\log(\omega)ω\omegalog⁡(ω)\log(\omega)ω0\omega^{0}

Table 1. Low-frequency behavior across different materials and geometries. In all configurations discussed within this work, the EELS probability Γ\Gamma diverges at least as ω−1\omega^{-1} in the low-frequency limit at finite temperatures. This behavior is inherited from the Bose–Einstein distribution nT(ω→0)∝ω−1n_T(\omega\to0)\propto\omega^{-1}. The low-frequency divergences are regularized in the frequency-resolved decoherence probability P(ω)P(\omega), yielding finite ω\omega-integrated values at finite interpath separations d⊥d_\perp.

P~ω(R,R′)=∫0ωdω′ P(R,R′,ω′),\widetilde{P}_{\omega}(\mathbf{R},\mathbf{R}')=\int_{0}^{\omega}d\omega'\,P(\mathbf{R},\mathbf{R}',\omega'),

which quantifies the decoherence associated with electrons that have gained or lost energy up to ℏω\hbar\omega during the interaction. This quantity is particularly relevant in experiments employing energy-filtered e-beams [21, 51], as it allows us to identify those inelastic channels that contribute maximally to electron decoherence. In practice, P~ω(R,R′)\widetilde{P}_{\omega}(\mathbf{R},\mathbf{R}') can be readily measured by inserting an energy filter between the interaction region and the electron detector or, ideally, by using a spectrometer. The filter could be introduced at any point after the interaction region, for example, just before the object plane z=z0z=z_0 (see Figure 1b). However, it may be more convenient to place it after beam recombination, for instance, after the second lens of the microscope. If a spectrometer is used, interferograms can be recorded on one axis of the camera, while the other axis resolves the electron kinetic energy. In this configuration, energy-filtered visibilities can be obtained by integrating the interferograms over a selected final energy range. For illustration, we focus on a two-path electron traversing a thin film (Figure 5a), as this configuration enhances temperature sensitivity owing to the higher-order divergence in the loss probability (see § 1.3 B and Table I).

Figure 5. Temperature sensing through decoherence. (a) A normally incident e-beam prepared in a two-path superposition intersects a thin film placed at a temperature TT. The transmitted e-beam is energy- and space-resolved at a detector, where the two paths are recombined. The spectrally resolved interference is used to measure the temperature of the sample. (b) Energy-filtered temperature sensitivity normalized to the corresponding fringe visibility as a function of energy-loss cutoff for 200 keV electrons traversing Al, Au, or LiF films of thickness L=100L=100 nm at T=300T=300 K or T=1000T=1000 K. We set the path separation to d⊥=50 μmd_\perp=50\,\mu\mathrm{m}. The dotted line indicates the bulk plasma frequency of Al. (c) Temperature dependence of the energy-filtered temperature sensitivity for a TT-dependent optimum energy-loss cutoff ℏωopt(T)\hbar\omega_{\mathrm{opt}}(T) (see inset). (d) Fringe visibility under the conditions of panel (c). The dotted line indicates T=300T=300 K.

To quantify the thermometric performance, we introduce the temperature sensitivity function, defined as the derivative of the fringe visibility with respect to temperature for the energy-filtered beam,

Sω(T)=∣∂Vω∂T∣=e−P~ω∣∂P~ω∂T∣,S_{\omega}(T)=\left|\frac{\partial V_{\omega}}{\partial T}\right|=e^{-\widetilde{P}_{\omega}}\left|\frac{\partial\widetilde{P}_{\omega}}{\partial T}\right|,

where we use the expression Vω=e−P~ω(d⊥)V_{\omega}=e^{-\widetilde{P}_{\omega}(d_\perp)} for the visibility [see Eq. (2)]. To improve performance, we choose a cutoff frequency ω\omega such that strong high-frequency features (e.g., plasmons in metals) are suppressed, while maintaining P~ω\widetilde{P}_{\omega} sufficiently large to remain measurable. Figure 5b shows SωS_{\omega} for Al, Au, and LiF films of 100 nm thickness. For all materials, within an energy-loss window between 100 meV and 10 eV, SωS_{\omega} reaches significantly higher values than those achieved when integrating over the full energy range. The lower bound to this energy window is determined by the interpath distance d⊥=50 μmd_{\perp}=50\,\mu\mathrm{m}, which imposes an effective cutoff on the wavelength of modes that can effectively imprint which-path information onto the environment and, thus, deplete interpath coherence. Interestingly, a large temperature sensitivity is observed for LiF, which is associated with phonon losses at ∼0.06\sim0.06 eV (i.e., close to kBTk_{\mathrm{B}}T). In conclusion, the sensitivity is optimized for energy cutoffs above optical phonons, but below the electronic band gap.

To formalize this idea, we define an optimal cutoff frequency ωopt(T)\omega_{\mathrm{opt}}(T) as the value of ω\omega that maximizes the temperature sensitivity Sω(T)S_{\omega}(T) at a fixed temperature TT. As shown in Figure 5c, the maximum sensitivity varies mildly for T≳200T \gtrsim200 K and coincides with the maxima identified in Figure 5b, predicting a nearly linear dependence of visibility on temperature. Indeed, as shown in Figure 5d, the visibility exhibits an approximately linear dependence on TT over most of the range considered, consistent with the low-frequency approximation nT(ω)≈kBT/ℏωn_{T}(\omega)\approx k_{\mathrm{B}}T/\hbar\omega together with the small-decoherence limit Vω≈1−P~ωV_{\omega}\approx1-\widetilde{P}_{\omega}. At ω=ωopt\omega=\omega_{\mathrm{opt}}, the visibility for metals and LiF changes by 10% and 2%, respectively. Finally, we note that increased sensitivity to low temperatures can be gained by increasing the interpath separation d⊥d_{\perp} in combination with a reduced energy cutoff.

Conclusions and outlook

In summary, we present a comprehensive study of the loss of spatial coherence experienced by free electrons due to interactions with material structures, which produce an impact-parameter-dependent entanglement of an e-beam with different excitations (i.e., a dependence on lateral position). Results are offered for two-path e-beams propagating inside a homogeneous material or near planar surfaces under parallel and perpendicular orientations. Decoherence should manifest in a loss of visibility upon recombination of the two paths, as illustrated above for a specific implementation suitable for electron microscopes. Different types of materials (e.g., plasmonic and phonon-polaritonic), spectral regions, and geometries are analyzed based on a unified formalism, offering a broad perspective of material-driven electron decoherence for general application in electron microscopy studies.

As a general rule, by recording all scattered electrons, we find that bulk excitations, such as bulk plasmons and interband transitions, play a dominant role in the depletion of coherence for metallic and dielectric materials, respectively, with only a weak dependence on temperature in the 0 – 1000 K range. This is in contrast to the contribution of surface modes. The temperature dependence of decoherence is governed by coupling to thermally populated low-frequency modes, such as free radiation, optical guided modes (i.e., Cherenkov emission), and long-lived surface plasmon polaritons: the population of those modes is enhanced by the infrared divergence in the Bose-Einstein distribution, which translates into a divergent EELS probability. This divergence is regularized in the decoherence probability, which incorporates the indistinguishability of the two paths when their separation is small compared to the wavelengths associated with the mode fields.

By examining different interaction configurations, we illustrate the roles played by various types of modes in the loss of coherence. For a split beam propagating entirely inside a bulk material at finite temperature, the emission of Cherenkov radiation leads to a non-divergent coupling at low frequencies, which is, however, rendered divergent by the thermal population at finite temperature. Interestingly, when the two electron paths propagate parallel to a planar material surface, decoherence is enhanced if one of the beams moves inside the medium and the other in the surrounding vacuum. This effect becomes particularly pronounced for materials such that the internal path couples to localized bulk excitations and the external path to surface modes. In this configuration, we find order-unity decoherence probabilities for relatively short electron path lengths (e.g., 50 nm in Al), implying a dramatic reduction in fringe visibility when the two paths are interfered. A low-frequency behavior of the loss probability analogous to that of the bulk configuration arises from both Cherenkov radiation inside the material and its radiative analogue outside.

We observe a more complex interplay of decoherence mechanisms when the two beams traverse a material film perpendicularly. In this configuration, the broad range of in-plane and out-of-plane momenta accessible to excitations multiplexes the inelastic channels responsible for decoherence by launching surface and waveguided modes and generating transition radiation. For conductive samples, this geometry produces an additional ω−1\omega^{-1} divergence in the energy-loss probability at zero temperature. At nonvanishing temperatures, this becomes a ω−2\omega^{-2} divergence, again regularized in the decoherence probability (see Table I). Overall, this configuration exhibits the strongest temperature dependence among the studied geometries. For example, while the variation of the decoherence probability PP over the 0 – 1000 K range for Al films of 100 nm thickness is on the order of 1% of its T=0T=0 value, the relative variation in the bulk material is nearly one order of magnitude smaller (see Table II). In LiF films, there is no zero-temperature divergence, and guided-mode excitations near the bulk phonon resonance in the meV range give rise to a modest temperature dependence.

MaterialLiFAl
Temperature (KK)300 K600 K900 K300 K600 K900 K
VωoptV_{\omega_{\mathrm{opt}}} (%)d⊥=5 μmd_{\perp}=5\ \mu\mathrm{m}99.298.798.299.498.998.4
d⊥=50 μmd_{\perp}=50\ \mu\mathrm{m}99.198.697.995.190.786.5
SωoptS_{\omega_{\mathrm{opt}}} (% K−1K^{-1})d⊥=5 μmd_{\perp}=5\ \mu\mathrm{m}0.00140.00170.00180.00070.00110.0012
d⊥=50 μmd_{\perp}=50\ \mu\mathrm{m}0.00160.00200.00210.01380.01380.0134

Table II. Fringe visibility and sensitivity to decoherence under different configurations. We provide representative values of the visibility VV and sensitivity SS for different path separations d⊥d_{\perp} and temperatures under the conditions of Figure 5 for LiF and Al films of 100 nm thickness. Results are presented for a TT-dependent energy-loss cutoff ℏωopt\hbar\omega_{\mathrm{opt}} as described in Figure 5.

Another important consequence of a higher-order divergence is the enhanced interpath dependence of decoherence, stemming from its connection to the cutoff wavelength at which the two paths become indistinguishable. In this context, we find that, for Al, the loss of interference fringes increases linearly with the separation between the two paths, whereas LiF films display a slower, logarithmic increase. This behavior is clearly captured by the ratio between surface and bulk contributions to decoherence (see Figure 4e), as well as by the corresponding values of visibility and temperature sensitivity reported in Table II. These findings are consistent with those predicted for a semi-infinite perfect-conductor plate [18], where an inverse-frequency spectral divergence yields a logarithmic increase in decoherence with path separation, while an inverse-square dependence results in linear growth.

The strong temperature dependence of the low-frequency energy-loss region in the perpendicular e-beam configuration suggests the study of decoherence as an approach to temperature sensing. To maximize this dependence, energy-filtered interference patterns can be recorded, allowing us to suppress temperature-independent contributions arising from bulk plasmons and electron-hole excitations in the 1–50 eV spectral region. We show that the spectral filtering window can be tuned to optimize temperature sensitivity. We find that collecting electrons with energies at distances from the zero-loss peak well below the plasmon energy yields an interference visibility of ∼97%\sim97\%, which in turn results in measurable changes of order 0.1% near room temperature across a 6 K variation in an Al film.

State-of-the-art nanoscale thermometry in transmission electron microscopy achieves combined thermal and spatial resolutions on the order of K-nm [52, 53, 54, 55, 56, 57, 58], with recent extensions to nanosecond temporal resolution at the cost of reduced thermal precision [59]. However, these approaches are typically material-specific and rely on specialized instrumentation. For instance, methods based on plasmon energy shifts [52, 53, 58, 59] track lattice thermal expansion, limiting their applicability to materials with well-defined plasmonic features and sizable thermal expansion coefficients. Techniques based on phonon population ratios in energy-loss and energy-gain spectra [54, 55, 59] require highly monochromated e-beams with energy spreads of only a few tens of meV. Cathodoluminescence-based methods further depend on temperature-sensitive emission features and an additional optical collection module [56, 57].

In contrast, our proposed approach applies to any conductive sample, as it exploits the high coupling probability to low-frequency radiative and surface plasmon modes that arise for penetrating electron trajectories at dielectric-conductor interfaces. The required instrumentation is limited to energy filtering as well as a sufficient transverse spatial coherence to enable coherent beam splitting into two paths separated by tens of microns. We also require structure sizes larger than the path-to-path separation to support the required long-wavelength modes. Under these conditions, temperature sensitivities on the order of 0.1% K−1K^{-1} (see Table II) can be achieved with a relatively weak material dependence (see Figure 5).

Beyond potential applications, the theoretical formulation of decoherence probabilities in terms of the electromagnetic Green tensor and the spectral decomposition of energy-loss channels provides a natural framework for generalization to more complex geometries, combining analytical insight with numerical methods. While our analysis focuses on electrons prepared in a spatial superposition of two point-like paths in the transverse plane, the formalism can be straightforwardly extended to configurations in which initially collimated beams are modified by a combination of elastic and inelastic scattering processes, leading to distortions of the full spatial structure of the electron density matrix (see Appendix D). In such scenarios, decoherence manifests not only as a reduction of fringe visibility but, more generally, as a redistribution of coherence across spatial modes. This perspective opens the door to the use of advanced reconstruction algorithms aimed at retrieving the reduced electron density matrix from experimentally accessible observables such as diffraction intensities [60]. In particular, ptychographic approaches [61] adapted to partially coherent e-beams [62, 63] offer a promising route to reconstruct both the amplitude and phase of the electron density matrix, including its off-diagonal elements, which directly encode decoherence. Such reconstruction schemes could enable quantitative extraction of temperature-dependent decoherence signatures, thereby providing an alternative and potentially powerful route to nanoscale thermometry based on full density-matrix reconstruction.

ACKNOWLEDGEMENTS

We thank Archie Howie and Jo Verbeeck for stimulating and enjoyable discussions. This work has been supported in part by the European Research Council (Adv. Grant 101141220-QUEFES), the European Commission (FET-Proactive 101017720-eBEAM), the Spanish MICIU (PID2024-157421NB-I00 and Severo Ochoa CEX2024-001490-S), and the CERCA Program.

Derivation of the intensity profile in Figure 1

To describe the propagation of an e-beam in the configuration shown in Figure 1b, we assume the two paths to be wide enough as to neglect diffraction over the entire setup (see main text). In addition, we assume axially-symmetric Gaussian profiles, such that each path j=A,Bj=A,B has an associated transverse wavefunction ψ⊥(R−Rj)\psi_\perp(\mathbf{R}-\mathbf{R}_j), where

ψ⊥(R)=2π3/4w0e−R2/w02(28)\psi_\perp(\mathbf{R})=\frac{\sqrt{2}}{\pi^{3/4}w_0}e^{-R^2/w_0^2} \tag*{(28)}

is an intensity-normalized component with a small Gaussian radius w0w_0 compared to the interpath separation ∣RA−RB∣\lvert\mathbf{R}_A-\mathbf{R}_B\rvert. The initial electron density matrix before interaction is ρ(R,z,R′,z)=(1/2)∑j,j′=A,Bψ⊥(R−Rj)ψ⊥(R′−Rj′).\rho(\mathbf{R},z,\mathbf{R}',z)=(1/2)\sum_{j,j'=A,B}\psi_\perp(\mathbf{R}-\mathbf{R}_j)\psi_\perp(\mathbf{R}'-\mathbf{R}_{j'}). Note that the complete sub-beam wavefunctions depend on the longitudinal coordinate zz through a propagation phase, which disappears in the product inside the density matrix when this is evaluated at the same zz plane.

The interaction is introduced through Eq. (1). We assume that neither the decoherence probability PP nor the elastic phase χ\chi varies significantly across the transverse extent of the two paths, and thus, they can be evaluated at the path centers as

e−P(R,R′)+iχ(R,R′)ψ⊥(R−Rj,z0)ψ⊥(R′−Rj′,z0)≈e−P(Rj,Rj′)+iχ(Rj,Rj′)ψ⊥(R−Rj,z0)ψ⊥(R′−Rj′,z0).e^{-P(\mathbf{R},\mathbf{R}')+i\chi(\mathbf{R},\mathbf{R}')}\psi_\perp(\mathbf{R}-\mathbf{R}_j,z_0)\psi_\perp(\mathbf{R}'-\mathbf{R}_{j'},z_0)\approx e^{-P(\mathbf{R}_j,\mathbf{R}_{j'})+i\chi(\mathbf{R}_j,\mathbf{R}_{j'})}\psi_\perp(\mathbf{R}-\mathbf{R}_j,z_0)\psi_\perp(\mathbf{R}'-\mathbf{R}_{j'},z_0).

Under this approximation, the entire beam propagation can be reduced to the independent evolution of each path. After the interaction, the corresponding density matrix becomes

ρ(R,z,R′,z)=12∑j,j′=A,Bψ⊥(R−Rj)ψ⊥(R′−Rj′)e−P(Rj,Rj′)+iχ(Rj,Rj′),(29)\rho(\mathbf{R},z,\mathbf{R}',z)=\frac{1}{2}\sum_{j,j'=A,B}\psi_\perp(\mathbf{R}-\mathbf{R}_j)\psi_\perp(\mathbf{R}'-\mathbf{R}_{j'})e^{-P(\mathbf{R}_j,\mathbf{R}_{j'})+i\chi(\mathbf{R}_j,\mathbf{R}_{j'})}, \tag*{(29)}

which includes path-dependent interaction factors.

We now introduce an electrostatic deflection of the sub-beams, which brings them together at an overlapping region in a plane z=z0z=z_0 placed a distance d∥d_\parallel from the mirror (see Figure 1b). Assuming a small deflection angle (i.e., d∥≫Rjd_\parallel\gg R_j) that maintains paraxial propagation conditions, the sub-beam wavefunctions become e−iq0Rj⋅R/d∥ψ⊥(R)e^{-i q_0\mathbf{R}_j\cdot\mathbf{R}/d_\parallel}\psi_\perp(\mathbf{R}) at the z0z_0 plane, up to a global position-independent phase factor. In addition, a two-lens microscope is introduced to produce an enlarged image of the z0z_0 plane at the conjugate, electron-detector plane ziz_i. For the parameters shown in Figure 1b, the magnification factor is given by the focal-length ratio g=f2/f1g=f_2/f_1, under the assumption that the lenses are separated by a distance f1+f2f_1+f_2 and the condition g2=−(d1−f1)/(d2−f2)g^2=-(d_1-f_1)/(d_2-f_2) is fulfilled, where d1d_1 and d2d_2 are the distances of the z0z_0 and ziz_i planes to lenses 1 and 2, respectively. The sub-beam wavefunctions at ziz_i become eiqj⋅rψ⊥(−R/g)e^{i\mathbf{q}_j\cdot\mathbf{r}}\psi_\perp(-\mathbf{R}/g), where qj=(q0/gd∥)Rj\mathbf{q}_j=(q_0/gd_\parallel)\mathbf{R}_j [64]. Performing a separate propagation of each path through this system and considering the trace of the density matrix [Eq. (A2), the detected electron intensity is given by ρ(R,zi,R,zi)=(1/2)∣ψ⊥(−R/g)∣2∑j,j′ei(qj−qj′)⋅Re−P(Rj,Rj′)+iχ(Rj,Rj′)\rho(\mathbf{R},z_i,\mathbf{R},z_i)=(1/2)\lvert\psi_\perp(-\mathbf{R}/g)\rvert^2\sum_{j,j'}e^{i(\mathbf{q}_j-\mathbf{q}_{j'})\cdot\mathbf{R}}e^{-P(\mathbf{R}_j,\mathbf{R}_{j'})+i\chi(\mathbf{R}_j,\mathbf{R}_{j'})}. Finally, inserting Eq. (28) into this expression, we readily obtain Eq. (2).

Lindhard and Mermin dielectric functions

We describe nonlocal effects in bulk metals by using the Lindhard dielectric function obtained in the RPA for a free-electron gas [43, 65]. For longitudinal fields, it is given by [65]

ϵL(q,ω)=1+2mee2kFπℏ2q2[1+R(q2kF,ωϵF)+R(q2kF,−ωϵF)].\epsilon_{\mathrm{L}}(q,\omega)=1+\frac{2m_e e^2 k_{\mathrm{F}}}{\pi\hbar^2q^2}\left[1+R\left(\frac{q}{2k_{\mathrm{F}}},\frac{\omega}{\epsilon_{\mathrm{F}}}\right)+R\left(\frac{q}{2k_{\mathrm{F}}},-\frac{\omega}{\epsilon_{\mathrm{F}}}\right)\right].

where ℏϵF=ℏ2kF2/2me\hbar\epsilon_{\mathrm{F}}=\hbar^2 k_{\mathrm{F}}^2/2m_e is the Fermi energy, kFk_{\mathrm{F}} is the Fermi wavenumber, and

R(x,y)=12x[1−(x2+y)24x2]log⁡(x2+2x+yx2−2x+y).R(x,y)=\frac{1}{2x}\left[1-\frac{(x^2+y)^2}{4x^2}\right]\log\left(\frac{x^2+2x+y}{x^2-2x+y}\right).

An inelastic damping rate η\eta cannot be simply introduced in this dielectric function through the ω→ω+iη\omega\to\omega+i\eta substitution because it breaks the conservation of local electron density. Instead, we adopt Mermin’s prescription to enforce such conservation, leading to the dielectric function [44]

ϵlonM(q,ω)=1+(ω+iη)[ϵL(q,ω+iη)−1]ω+iη[ϵL(q,ω+iη)−1]/[ϵL(q,0)−1].\epsilon_{\mathrm{lon}}^{\mathrm{M}}(q,\omega)=1+\frac{(\omega+i\eta)[\epsilon_{\mathrm{L}}(q,\omega+i\eta)-1]}{\omega+i\eta[\epsilon_{\mathrm{L}}(q,\omega+i\eta)-1]/[\epsilon_{\mathrm{L}}(q,0)-1]}.

In the local approximation, the Mermin dielectric function reduces to the Drude model ϵlonM(q→0,ω)=ϵD(ω)\epsilon_{\mathrm{lon}}^{\mathrm{M}}(q\to0,\omega)=\epsilon^{\mathrm{D}}(\omega), see Eq. (19). For the transverse component, we also use the Drude model, as nonlocal corrections are negligible for metals such as Al and Au [65].

Induced field for an electron crossing a thin film

As mentioned in § 1.3 B, the coefficients for the induced field in Eq. (26) are obtained by imposing the continuity of the parallel components of the electric and magnetic fields at the z=0z=0 and z=Lz=L surface planes. For the electric field, we directly use Eqs. (25) and (26). For the magnetic field, we obtain the bulk form in a homogeneous medium jj (i.e., either the material or the external vacuum) by applying Faraday’s law to Eq. (25), leading to

Hbulk(Q,z,ω)=4πieQeiωz/vc(q2−k2ϵj)eQs\mathbf{H}^{\mathrm{bulk}}(\mathbf{Q},z,\omega)=\frac{4\pi i e Q e^{i\omega z/v}}{c(q^2-k^2\epsilon_j)}\mathbf{e}_{Q_s}

with eQs=(−qyx^+qxy^)/Q\mathbf{e}_{Q_s}=(-q_y\hat{\mathbf{x}}+q_x\hat{\mathbf{y}})/Q, while the surface contribution is obtained from Eq. (26):

Hsurf(Q,z,ω)=eQs{Ae−iqz0z,z<0,ϵj∑±C±e±iqz1z,0<z<L,Beiqz0z,z>L.\mathbf{H}^{\mathrm{surf}}(\mathbf{Q},z,\omega)=\mathbf{e}_{Q_s}\begin{cases} A e^{-iq_{z0}z}, & z<0,\\ \sqrt{\epsilon_j}\sum_{\pm}C_{\pm}e^{\pm iq_{z1}z}, & 0<z<L,\\ B e^{iq_{z0}z}, & z>L. \end{cases}

Note that both bulk and surface magnetic fields are ss-polarized. Projecting the electric and magnetic fields over Q\mathbf{Q} and eQs\mathbf{e}_{Q_s}, respectively, yields a linear system of equations for the unknown coefficients AA, BB, and C±C_{\pm}, which can be written in compact matrix form as M⋅C=b\mathcal{M}\cdot\mathbf{C}=\mathbf{b}, where C=(A,B,C+,C−)T\mathbf{C}=(A,B,C_{+},C_{-})^{\mathsf{T}},

M=(10−ϵ−ϵ0−eiqz0Lξ1ϵξ1−1ϵqz0/k0qz1/kϵ−qz1/kϵ0ξ0qz0/k−ξ1qz1/kϵξ1−1qz1/kϵ)\mathcal{M}=\begin{pmatrix} 1 & 0 & -\sqrt{\epsilon} & -\sqrt{\epsilon}\\ 0 & -e^{iq_{z0}L} & \xi_1\sqrt{\epsilon} & \xi_1^{-1}\sqrt{\epsilon}\\ q_{z0}/k & 0 & q_{z1}/k\sqrt{\epsilon} & -q_{z1}/k\sqrt{\epsilon}\\ 0 & \xi_0q_{z0}/k & -\xi_1q_{z1}/k\sqrt{\epsilon} & \xi_1^{-1}q_{z1}/k\sqrt{\epsilon} \end{pmatrix}

and b=(b1,−b1eiωL/v,b2,−b2eiωL/v)T\mathbf{b}=(b_1,-b_1e^{i\omega L/v},b_2,-b_2e^{i\omega L/v})^{\mathsf{T}}, with ξj=eiqzjL\xi_j=e^{iq_{zj}L}, b1=C0(ϵ−1)b_1=C_0(\epsilon-1), b2=C0(1/ϵ−1)(c/v)(1+ϵ−q2/k2)b_2=C_0(1/\epsilon-1)(c/v)(1+\epsilon-q^2/k^2), and C0=4πieQ(k2/c)/[(q2−k2)(q2−k2ϵ)]C_0=4\pi i e Q(k^2/c)/[(q^2-k^2)(q^2-k^2\epsilon)]. The solution

C=M−1⋅b\mathbf{C}=\mathcal{M}^{-1}\cdot\mathbf{b}

is explicitly given by

C+=C0(ϵ−1)tp01[κ+2+eiωL/vξ1rp01κ−2]F−1,C_{+}=C_0(\epsilon-1)t_p^{01}\left[\kappa_{+}^{2}+e^{i\omega L/v}\xi_1r_p^{01}\kappa_{-}^{2}\right]F^{-1},
C−=e2iqz1LC0(1−ϵ)tp01[κ−2eiωL/vξ1−1+rp01κ+2]F−1,C_{-}=e^{2iq_{z1}L}C_0(1-\epsilon)t_p^{01}\left[\kappa_{-}^{2}e^{i\omega L/v}\xi_1^{-1}+r_p^{01}\kappa_{+}^{2}\right]F^{-1},
A=ϵ(C++C−)+C0(ϵ−1),A=\sqrt{\epsilon}(C_{+}+C_{-})+C_0(\epsilon-1),
B=ξ0−1{ϵ[C+ξ1+C−ξ1−1]+C0(ϵ−1)eiωL/v},B=\xi_0^{-1}\left\{\sqrt{\epsilon}\left[C_{+}\xi_1+C_{-}\xi_1^{-1}\right]+C_0(\epsilon-1)e^{i\omega L/v}\right\},

where F=2ϵ(vk/c)qz0[1−(rp01ξ1)2]F=2\epsilon(vk/c)q_{z0}\left[1-(r_p^{01}\xi_1)^2\right] and κ±2=q2−k2[1+ϵ±(qz0v/ω)ϵ]\kappa_{\pm}^{2}=q^2-k^2[1+\epsilon\pm(q_{z0}v/\omega)\epsilon].

Propagation of arbitrary transverse e-beam profiles

Although we focus the present work on two-path e-beams for tutorial purposes, the formalism in Sec. II can be applied to any transverse profile. Propagation from a post-interaction plane z=z0z=z_0 to a subsequent plane z=z1z=z_1 along the e-beam column simply mixes different transverse components of the density matrix in Eq. (1). By moving from transverse coordinates R\mathbf{R} and R′\mathbf{R}' to transverse momentum space Q\mathbf{Q} and Q′\mathbf{Q}' according to ρ(Q,z0,Q′,z0)=∫d2R∫d2R′e−i(Q⋅R−Q′⋅R′)ρ(R,z0,R′,z0)\rho(\mathbf{Q},z_0,\mathbf{Q}',z_0)=\int d^2\mathbf{R}\int d^2\mathbf{R}'e^{-i(\mathbf{Q}\cdot\mathbf{R}-\mathbf{Q}'\cdot\mathbf{R}')}\rho(\mathbf{R},z_0,\mathbf{R}',z_0), the propagated state can be written

ρ(R,z1,R′,z1)=∫d2Q(2π)2∫d2Q′(2π)2ρ(Q,z0,Q′,z0)ei(Q⋅R−Q′⋅R′)+i(qz−qz′)(z1−z0)(30)\rho(\mathbf{R},z_1,\mathbf{R}',z_1)=\int\frac{d^2\mathbf{Q}}{(2\pi)^2}\int\frac{d^2\mathbf{Q}'}{(2\pi)^2}\rho(\mathbf{Q},z_0,\mathbf{Q}',z_0)e^{i(\mathbf{Q}\cdot\mathbf{R}-\mathbf{Q}'\cdot\mathbf{R}')+i(q_z-q_z')(z_1-z_0)} \tag*{(30)}

where qz=q0−Q2q_z=\sqrt{q_0-Q^2} and qz′=q0−Q′2q_z'=\sqrt{q_0-Q'^2} are longitudinal electron wavevector components and q0q_0 is the total wavenumber. Under the paraxial approximation (qz≈q0−Q2/2q0q_z\approx q_0-Q^2/2q_0), the Q\mathbf{Q} and Q′\mathbf{Q}' integrals can be performed analytically to recast Eq. (30) into a convolution as [64]

ρ(R,z1,R′,z1)=q024π2(z1−z0)2∫d2R′′∫d2R′′′ρ(R′′,z0,R′′′,z0)eiq0(∣R−R′′∣2−∣R′−R′′′∣2)/2(z1−z0).\rho(\mathbf{R},z_1,\mathbf{R}',z_1)=\frac{q_0^2}{4\pi^2(z_1-z_0)^2}\int d^2\mathbf{R}''\int d^2\mathbf{R}'''\rho(\mathbf{R}'',z_0,\mathbf{R}''',z_0)e^{iq_0(|\mathbf{R}-\mathbf{R}''|^2-|\mathbf{R}'-\mathbf{R}'''|^2)/2(z_1-z_0)}.

Combining this result with Eq. (1) explicitly demonstrates how the decoherence probability and the elastic phase jointly influence the evolution of the electronic state. In particular, for a fully coherent incident e-beam characterized by a transverse wavefunction ψ⊥(R,z0)\psi_{\perp}(\mathbf{R},z_0), the post-interaction density matrix is given by ρ(R,z0,R′,z0)=e−P(R,z0,R′,z0)eiχ(R,z0,R′,z0)ψ⊥(R,z0)ψ⊥∗(R′,z0)\rho(\mathbf{R},z_0,\mathbf{R}',z_0)=e^{-P(\mathbf{R},z_0,\mathbf{R}',z_0)}e^{i\chi(\mathbf{R},z_0,\mathbf{R}',z_0)}\psi_{\perp}(\mathbf{R},z_0)\psi_{\perp}^{*}(\mathbf{R}',z_0).

SUPPLEMENTARY FIGURES

Surface contribution to frequency-resolved decoherence probability

Figure S7. Zero-temperature limit: Surface contribution to electron decoherence: Parallel e-beam configuration Surface contribution to the frequency-resolved decoherence probability Psurf(ω)P^{\mathrm{surf}}(\omega) for a 200 keV electron under the configuration of Figure 3a in the main text when the material is Au, Al, or LiF at T=0T = 0. We set d0=d1=0.5 μmd_0 = d_1 = 0.5\,\mu\mathrm{m}. As PsurfP^{\mathrm{surf}} becomes negative for some frequencies, we plot both PsurfP^{\mathrm{surf}} (solid curves) and −Psurf-P^{\mathrm{surf}} (broken curves).

References

  1. [3]B. Wolter, M. G. Pullen, A.-T. Le, M. Baudisch, K. Doblhoff-Dier, A. Senftleben, M. Hemmer, C. D. Schröter, J. Ullrich, T. Pfeifer, R. Moshammer, S. Gräfe, O. Vendrell, C. D. Lin, and J. Biegert, “Ultrafast electron diffraction imaging of bond breaking in di-ionized acetylene,” Science 354, 308–312 (2016).pubmed.ncbi.nlm.nih.gov/27846561
  2. [4]D. Filippetto, P. Musumeci, R. K. Li, B. J. Siwick, M. R. Otto, M. Centurion, and J. P. Nunes, “Ultrafast electron diffraction: Visualizing dynamic states of matter,” Rev. Mod. Phys. 94, 045004 (2022).DOI
  3. [5]M. Lehmann and H. Lichte, “Tutorial on off-axis electron holography,” Microsc. Microanal. 8, 447–466 (2002).pubmed.ncbi.nlm.nih.gov/12533207
  4. [6]Florian Winkler, Juri Barthel, Amir H. Tavabi, Sven Borghardt, Beata E. Karydas, and Rafal E. Dunin-Borkowski, “Absolute scale quantitative off-axis electron holography at atomic resolution,” Phys. Rev. Lett. 120, 156101 (2018).DOI
  5. [7]J. R. Anglin, J. P. Paz, and W. H. Zurek, “Deconstructing decoherence,” Phys. Rev. A 55, 4041–4053 (1997).DOI
  6. [8]F. D. Mazzitelli, J. P. Paz, and A. Villanueva, “Decoherence and recoherence from vacuum fluctuations near a conducting plate,” Phys. Rev. A 68, 062106 (2003).DOI
  7. [9]Y. Levinson, “Decoherence of electron beams by electromagnetic field fluctuations,” J. Phys. A: Math. Gen. 37, 3003–3017 (2004).
  8. [10]J.-T. Hsiang and D.-S. Lee, “Influence of electron coherence from quantum electromagnetic fields in the presence of conducting plates,” Phys. Rev. D 73, 065022 (2006).DOI
  9. [11]P. Machnikowski, “Theory of which path dephasing in single electron interference due to trace in conductive environment,” Phys. Rev. B 73, 155109 (2006).DOI
  10. [12]Peter Sonnentag and Franz Hasselbach, “Measurement of decoherence of electron waves and visualization of the quantum-classical transition,” Phys. Rev. Lett. 98, 200402 (2007).pubmed.ncbi.nlm.nih.gov/17677675
  11. [13]F. Hasselbach, “Progress in electron- and ion-interferometry,” Rep. Prog. Phys. 73, 016101 (2010).DOI
  12. [14]P. J. Beierle, L. Zhang, and H. Batelaan, “Experimental test of decoherence theory using electron matter waves,” New J. Phys. 20, 113038 (2018).DOI
  13. [15]N. Kerker, R. Röpke, L. M. Steinert, A. Pooch, and A. Stibor, “Quantum decoherence by Coulomb interaction,” New J. Phys. 22, 063039 (2020).
  14. [16]L. H. Ford, “Electromagnetic vacuum fluctuations and electron coherence,” Phys. Rev. D 47, 5571–5580 (1993).pubmed.ncbi.nlm.nih.gov/10015584
  15. [17]L. H. Ford, “Electromagnetic vacuum fluctuations and electron coherence. II. Effects of wave-packet size,” Phys. Rev. A 56, 1812–1818 (1997).DOI
  16. [18]C. I. Velasco, V. Di Giulio, and F. J. García de Abajo, “Radiative loss of coherence in free electrons: a long-range quantum phenomenon,” Light Sci. Appl. 13, 31 (2024).DOI
  17. [19]V. Di Giulio and F. J. García de Abajo, “Electron diffraction by vacuum fluctuations,” New J. Phys. 22, 103057 (2020).DOI
  18. [20]H. Lichte and B. Freitag, “Inelastic electron holography,” Ultramicroscopy 81, 177–186 (2000).
  19. [21]R. A. Herring, “Energy-filtered electron-diffracted beam holography,” Ultramicroscopy 104, 261–270 (2005).pubmed.ncbi.nlm.nih.gov/15996821
  20. [22]P. L. Potapov, H. Lichte, J. Verbeeck, and D. van Dyck, “Experiments on inelastic electron holography,” Ultramicroscopy 106, 1012–1018 (2006).pubmed.ncbi.nlm.nih.gov/16934930
  21. [23]P. L. Potapov, J. Verbeeck, P. Schattschneider, H. Lichte, and D. van Dyck, “Inelastic electron holography as a variant of the Feynman thought experiment,” Ultramicroscopy 107, 559–567 (2007).pubmed.ncbi.nlm.nih.gov/17215083
  22. [24]J. Verbeeck, G. Bertoni, and P. Schattschneider, “The Fresnel effect of a defocused biprism on the fringes in inelastic holography,” Ultramicroscopy 108, 263–269 (2008).pubmed.ncbi.nlm.nih.gov/18036742
  23. [25]A. Howie, “Mechanisms of decoherence in electron microscopy,” Ultramicroscopy 111, 761–767 (2011).DOI
  24. [26]S. Scheel and S. Y. Buhmann, “Path decoherence of charged and neutral particles near surfaces,” Phys. Rev. A 85, 030101 (2012).DOI
  25. [27]Z. Chen and H. Batelaan, “Dephasing due to semiconductor charging masks decoherence in electron-wall systems,” Europhys. Lett. 129, 40004 (2020).DOI
  26. [28]W. Magnus, “On the exponential solution of differential equations for a linear operator,” Comm. Pure Appl. Math. VII, 649–673 (1954).DOI
  27. [29]F. J. García de Abajo, “Optical excitations in electron microscopy,” Rev. Mod. Phys. 82, 209–275 (2010).DOI
  28. [30]F. J. García de Abajo and V. Di Giulio, “Optical excitations with electron beams: Challenges and opportunities,” ACS Photonics 8, 945–974 (2021).DOI
  29. [31]F. J. García de Abajo and A. Howie, “Retarded field calculation of electron energy loss in inhomogeneous dielectrics,” Phys. Rev. B 65, 115418 (2002).DOI
  30. [32]A. A. Lucas, E. Kartheuser, and R. G. Badro, “Electron-phonon interaction in dielectric films. Application to electron energy loss and gain spectra,” Phys. Rev. B 2, 2488–2499 (1970).DOI
  31. [33]M. Šunjić and A. A. Lucas, “Multiple plasmon effects in the energy-loss spectra of electrons in thin films,” Phys. Rev. B 3, 719–729 (1971).DOI
  32. [34]F. J. García de Abajo, “Multiple excitation of confined graphene plasmons by single electrons,” ACS Nano 7, 11409–11419 (2013).
  33. [35]A. Imbusch and H. Niedrig, “Temperature effect on energy loss spectrum of fast electrons in aluminium and lead foils between 3 K and 295 K,” Phys. Lett. A 32, 375–376 (1970).DOI
  34. [36]Harsha Reddy, Urcan Guler, Alexander V. Kildishev, Alexandra Boltasseva, and Vladimir M. Shalaev, “Temperature-dependent optical properties of gold thin films,” Opt. Mater. Express 6, 2776–2802 (2016).DOI
  35. [37]M. Gottlieb, “Optical properties of lithium fluoride in the infrared,” J. Opt. Soc. Am. 50, 343–349 (1960).
  36. [38]J.R. Jasperse, A. Kahan, J.N. Plendl, and S.S. Mitra, “Temperature dependence of infrared dispersion in ionic crystals LiF and MgO,” Phys. Rev. 146, 526–542 (1966).
  37. [39]S. Tongay, J. Zhou, C. Ataca, K. Lo, T.S. Matthews, J. Li, J.C. Grossman, and J. Wu, “Thermally driven crossover from indirect toward direct bandgap in 2D semiconductors: MoSe₂ versus MoS₂,” Nano Lett. 12, 5576–5580 (2012).DOI
  38. [40]A.T. Hanbicki, M. Currie, G. Kioseoglou, A.L. Friedman, and B.T. Jonker, “Measurement of high exciton binding energy in the monolayer transition-metal dichalcogenides WS₂ and WSe₂,” Solid State Commun. 203, 16–20 (2015).DOI
  39. [41]M. Selig, G. Berghäuser, A. Raja, P. Nagler, C. Schüller, T.F. Heinz, T. Korn, A. Chernikov, E. Malic, and A. Knorr, “Excitonic linewidth and coherence lifetime in monolayer transition metal dichalcogenides,” Nat. Commun. 7, 13279 (2016).
  40. [42]E. D. Palik, Handbook of Optical Constants of Solids (Academic Press, San Diego, 1985).
  41. [43]J. Lindhard, “On the properties of a gas of charged particles,” K. Dan. Vidensk. Selsk. Mat. Fys. Medd. 28, 1–57 (1954).
  42. [44]N. D. Mermin, “Lindhard dielectric function in the relaxation-time approximation,” Phys. Rev. B 1, 2362–2363 (1970).DOI
  43. [45]P. B. Johnson and R. W. Christy, “Optical constants of the noble metals,” Phys. Rev. B 6, 4370–4379 (1972).DOI
  44. [46]F. J. García de Abajo, “Nonlocal effects in the plasmons of strongly interacting nanoparticles, dimers, and waveguides,” J. Phys. Chem. C 112, 17983–17987 (2008).arxiv.org/abs/0802.0040
  45. [48]F. J. García de Abajo, A. Rivacoba, N. Zabala, and N. Yamamoto, “Boundary effects in Cherenkov radiation,” Phys. Rev. B 69, 155420 (2004).DOI
  46. [49]S. Y. Buhmann, Dispersion Forces I. Macroscopic Quantum Electrodynamics and Ground-State Casimir, Casimir-Polder and van der Waals Forces (Springer-Verlag Berlin Heidelberg, Verlag Berlin Heidelberg, 2012).
  47. [50]V. L. Ginzburg and I. M. Frank, “Radiation of a uniformly moving electron due to its transition from one medium to another,” Zh. Eksp. Teor. Fiz. [Sov. Phys. JETP] 16, 15–28 (1946).
  48. [51]A. S. Eggeman, P. J. Dobson, and A. K. Petford-Long, “Optical spectroscopy and energy-filtered transmission electron microscopy of surface plasmons in core-shell nanoparticles,” J. Appl. Phys. 101, 024307 (2007).DOI
  49. [52]M. Mecklenburg, W. A. Hubbard, E. R. White, R. Dhall, S. B. Cronin, S. Aloni, and B. C. Regan, “Nanoscale temperature mapping in operating microelectronic devices,” Science 347, 629 (2015).DOI
  50. [53]A. Chmielewski, C. Ricciolleau, D. Alloyeau, G. Wang, and J. Nelayah, “Nanoscale temperature measurement during temperature controlled in situ TEM using Al plasmon nanothermometry,” Ultramicroscopy 209, 112881 (2020).DOI
  51. [54]M. J. Lagos and P. E. Batson, “Thermometry with sub-nanometer resolution in the electron microscope using the principle of detailed balancing,” Nano Lett. 18, 4556–4563 (2018).DOI
  52. [55]J. C. Idrobo, A. R. Lupini, T. Feng, R. R. Unocic, F. S. Walden, D. S. Gardiner, T. C. Lovejoy, N. Delby, S. T. Pantelides, and O. L. Krivanek, “Temperature measurement by a nanoscale electron probe using energy gain and loss spectroscopy,” Phys. Rev. Lett. 120, 095901 (2018).
  53. [56]K. W. Mauser, M. Solà-Garcia, M. Lietbrau, B. Damilano, P.-M. Coulon, P. A. Shields S. Vézian, S. Meuret, and A. Polman, “Employing cathodoluminescence for nanothermometry and thermal transport measurements in semiconductor nanowires,” ACS Nano 18, 4911–4921 (2024).
  54. [57]W.-W. Park, P. K. Olshin, Y.-J. Kim, H.-W. Nho, D. V. Mamonova, I. E. Kolesnikov, V. A. Medvedev, and O.-H. Kwon, “Nanoscale cathodoluminescence thermometry with a lanthanide-doped heavy-metal oxide in transmission electron microscopy,” ACS Nano 18, 4911–4921 (2024).DOI
  55. [58]L. D. Palmer, W. Lee, D. B. Durham, Jr. J. Fajardo, Y. Liu, A. A. Talin, T. E. Gage, and S. K. Cushing, “Nanoscale and element-specific lattice temperature measurements using core-loss electron energy-loss spectroscopy,” ACS Phys. Chem. Au 5, 589–598 (2025).DOI
  56. [59]F. Castioni, Y. Auad, J. D. Blazit, X. Li, S. Y. Woo, K. Watanabe, T. Taniguchi, C.-H. Ho, O. Stéphann, M. Kociak, and L. H. G. Tizei, “Nanoscond nanothermometry in an electron microscope,” Nano Lett. 25, 1601–1608 (2025).DOI
  57. [60]S. Cao, P. Kok, A. M. Maiden, and J. M. Rodenburg, “Modal decomposition of a propagating matter wave via electron ptychography,” Phys. Rev. A 94, 063621 (2016).DOI
  58. [61]H. L. Lalande Robert, M. L. Leidl, K. Müller-Caspary, and J. Verbeeck, “Benchmarking analytical electron ptychography methods for the low-dose imaging of beam-sensitive materials,” Eur. Phys. J. Appl. Phys. 100, 20 (2025).
  59. [62]Z. Chen, M. Odstrcil, Y. Jiang, Y. Han, M.-H. Chiu, L.-J. Li, and D. A. Muller, “Mixed-state electron ptychography enables sub-angstrom resolution imaging with picometer precision at low dose,” Nat. Commun. 11, 2994 (2020).DOI
  60. [63]W. Xu, S. Ning, P. Sheng, H. Lin, A. I. Kirkland, and Y. Peng, “A high-performance reconstruction method for partially coherent ptychography,” Ultramicroscopy 267, 114068 (2024).arxiv.org/abs/2406.06640
  61. [64]F. J. García de Abajo and C. I. Velasco, “Spectrometer-free electron spectroscopy,” ACS Nano 19, 35770–35776 (2025).DOI
  62. [65]Martin Dressel and George Grüner, Electrodynamics of Solids: Optical Properties of Electrons in Matter (Cambridge University Press, Cambridge, 2002).

Paper details

vlnc:W_4wzvtvv7
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