WorksPhysics and Astronomy
Coulomb Crystallization in Free Electron Beams
Abstract
Classical Coulomb crystals and quantum Wigner crystals typically arise when repulsive interactions order like-charged particles in spatially confined environments. Here, we show that transient crystallization can also emerge in freely propagating electron pulses without external trapping. Experimentally, we use a transmission electron microscope to measure number-resolved energy correlations in photoemitted pulses containing up to nine electrons, revealing strong correlations at small particle numbers that progressively disappear in the pulse interior as the number increases. To explain these observations, we combine many-body simulations with a one-dimensional statistical-mechanical model, showing that the leading and trailing electrons generate a self-induced longitudinal potential that transiently confines the inner particles, enabling the formation of an ordered chain during propagation. The resulting crystallization phase diagram reveals how many-body ordering dynamics can be understood in terms of evolving macroscopic beam properties, namely its geometry and the ratio of the mean Coulomb interaction energy to the longitudinal thermal energy. Extending this framework, we theoretically demonstrate that weak transverse guiding through electrostatic or optical ponderomotive forces can sustain crystallization at larger particle numbers and predict spontaneous longitudinal periodic bunching in continuous electron beams. Our results identify conditions under which stochastic Coulomb interactions can be harnessed to generate self-structured, strongly correlated electron beams in both pulsed and continuous regimes.
Introduction
Long-range interactions can govern the behavior of many-body systems, promoting the spatial stabilization of their components and giving rise to ordered phases in the limit of strong coupling. For like-charged particles, the equilibrium configuration takes the form of a crystalline structure when the average Coulomb energy , arising from repulsive inter-particle interactions, dominates over the average kinetic energy [1]. In classical systems, where the kinetic energy is set by thermal motion, this condition can be achieved by increasing the particle density or by lowering the temperature. Due to the repulsive nature of the interaction, Coulomb-induced crystallization generally requires the presence of a confining potential. This has so far been realized through the external confinement of laser-cooled ions in Paul traps [2] or through alternating static electric and magnetic multipoles in storage rings [3, 4, 5, 6], giving rise to so-called Coulomb crystals [1, 7, 8]. In quantum systems, in which the Fermi energy sets the relevant kinetic-energy scale, decreasing the density can instead drive a quantum transition toward spatial ordering [9]. This regime is typically reached in cryogenically cooled solid-state systems, where confinement is naturally provided by the atomic potential landscape, allowing conduction electrons to crystallize. Such ordered structures are known as Wigner crystals [10] and have been observed in carbon nanotubes [11, 12], in twisted transition-metal dichalcogenides [13], and twisted bilayer graphene [14]. Beyond equilibrium crystalline phases, Coulomb-induced correlations have been observed in electron pairs exhibiting interaction-driven antibunching and partitioning when traveling in cryogenic semiconductor nanos-
structures, including ballistic edge-channel beam splitters [15, 16] and surface-acoustic-wave circuits [17]. Moving few-electron droplets in a GaAs heterostructure have further revealed signatures of a correlated Coulomb liquid [18].
Electron microscopes inherently rely on beams of equally charged particles that are focused onto a specimen, with atomic-resolution images and spectroscopic information reconstructed from the properties of the transmitted and scattered electrons [19–21]. Precise electron microscopy and spectroscopy therefore require electron beams (e-beams) with tightly controlled distributions in energy, space, and time, such that scattering-induced changes can be distinguished from the intrinsic spread of the incident beam. In this context, Coulomb repulsion has traditionally been regarded as detrimental because it gives rise to the Boersch [22–24] and Löffler effects [25, 26] and, more generally, increases the occupied phase-space volume while reducing beam brightness [26, 27]. More recently, however, Coulomb-driven electron-electron interactions in free space have attracted interest as a means of forming well-defined correlations in the spatial, temporal, and energy degrees of freedom of few-electron pulses photoemitted from Schottky field emitters. Early evidence for such interaction-induced correlations was found in the form of arrival-time anticorrelations and sub-Poissonian number statistics, observed for ultrashort photoelectron pulses from nanotips [28], spin-polarized beams from semiconductor photocathodes [29], and continuous field-emission beams probed with sub-picosecond resolution [30]. Their formation has since been demonstrated in transmission electron microscopes [31, 32] and from freestanding nanotips [33, 34], and they have been characterized via delayed near-field excitation in the time [32, 35], energy [32, 36], and energy-time domains [37]. Until now, their formation has been limited to electrons per pulse, while the conditions required to extend these correlations to larger electron numbers remain not fully understood, thus hindering further advances toward the synthesis of strongly correlated e-beams at higher .
Overcoming this limitation would open the possibility of harnessing these correlations at high currents. Combined with spatial and energy filtering, this could enable the generation of sub-shot-noise e-beams [38, 39] with deterministic and controlled phase-space properties [39]. Furthermore, Coulomb crystallized e-beams spontaneously bunch in time, similarly to charged particles in plasma wakefields [40–43]. Such bunching may in turn give rise to coherent superradiant emission [44], akin to that generated in free-electron lasers [45]. Indeed, as demonstrated for ions [5], collective interactions can lead to the formation of short pulses without external compression fields. Applied to pulsed and continuous electron emission, this approach could improve control over particle-number and current fluctuations, benefiting electron-beam lithography [46], microscopy, and spectroscopy through enhanced dose stability, precision, and sensitivity [47–49].
In this work, we demonstrate that electron pulses photoemitted from a metallic nanotip can form free-space Coulomb crystals, confined not by an external trap but by a potential generated by the pulse itself. We show that, in one-dimensional beams with high pulse current, the leading and trailing electrons form a self-induced potential that confines the inner particles in free space. Such an emergence of order stands in stark contrast to previously studied Coulomb crystals in storage rings, moiré superlattices, and carbon nanotubes (see Figure 1a), where order formation fundamentally relies on external trapping forces. By combining Monte Carlo simulations with experimental measurements of -electron beams (Ne-beams) with 4–9 particles per pulse in a transmission electron microscope, we determine the phase diagram of a one-dimensional chain model and demonstrate that mechanically and thermally stable crystalline configurations can form in high-aspect-ratio Ne-beams. A transition from spatial order to disorder is observed with increasing due to a decrease in the time spent in such a one-dimensional configuration for the electrons to order through intra-beam scattering.

Figure 1. Formation of self-trapped Coulomb crystals. (a) Overview of selected systems in which Coulomb crystals have been observed, shown as a function of the Coulomb coupling, i.e., the ratio between Coulomb energy and kinetic energy . These include externally trapped classical systems, such as ions in storage rings [3, 4, 5, 6], and quantum Coulomb crystals (Wigner crystals), such as electrons in carbon nanotubes [11] [50] and moiré superlattices in two-dimensional heterostructures [13]. As demonstrated in this work, -electron beams (-beams) photoemitted from a metallic nanotip in an electron microscope can also crystallize under their self-induced potential. (b) During propagation, a photoemitted -beam with longitudinal extension and lateral size evolves from a large aspect ratio , where leading and trailing electrons generate a strong self-trapping potential (blue lines), to a smaller aspect ratio due to beam divergence and Coulomb expansion. The interplay between Coulomb repulsion and self-induced trapping orders the electrons within the pulse, producing a Coulomb crystal visible in the electron density (dark blue). (c) Phase diagram obtained from the one-dimensional softened-Coulomb model for a 5-electron pulse with fixed leading and trailing electrons and fluctuating inner electrons, plotted as a function of aspect ratio and Coulomb coupling , with . The solid-disordered transition line is determined from the Lindemann criterion (see Eq. (1)), while the dashed black line indicates the minimum aspect ratio required for mechanical stability of the chain.
We then numerically show that such loss of order can be overcome by introducing a weak transverse confinement between the extractor and linear accelerator that preserves a high pulse aspect ratio, thus sustaining correlations at larger . Our simulations predict that even a single refocusing event enhances the final electron-electron correlations relative to the case with no trapping, while a longer millimeter-scale trapping stage can produce crystalline pulses with larger particle numbers, reaching up to electrons per pulse under the same experimental conditions. Finally, using molecular dynamics simulations, we extend the same principle to transversely confined continuous beams, i.e., to the limit . We show that this equilibrium can exhibit strong longitudinal bunching and Coulomb ordering, with universal properties controlled by a normalized linear electron density and the Coulomb coupling. By linking these quantities to the input-beam properties and trapping parameters through energy conservation arguments, we theoretically establish the conditions for realizing continuous Coulomb-ordered free-electron beams in electron microscopes.
Crystallization of free N-electron pulses
In an ultrafast transmission electron microscope, pulsed beams are generated by photoemission from a metallic nanotip [51] (see Figure 1b) and subsequently accelerated by an extractor electrode [31, 37]. During propagation along the microscope column, the beam configuration evolves under the combined effects of the accelerating fields, its initial divergence, dispersion, and Coulomb-driven expansion, which modify properties such as the beam aspect ratio. Close to the tip, the finite duration of photoemission gives the electrons relative emission delays, and the strong extraction field then produces an initial axial stretching that renders the pulse effectively one dimensional. Its approximate geometry can be determined by the initial transverse beam size , the semi-angle spread , and the extractor potential . For instance, at the extractor plane, located a distance from the tip, the longitudinal and transverse dimensions of a non-interacting beam are and , respectively. For the experimental setup, typical values are fs, kV, mrad, nm, and m, these expressions yield an aspect ratio and a linear density m.
The combination of a large aspect ratio and high linear density provides the physical conditions for the emergence of a spatially ordered electron phase without external transverse confinement. In the quasi-one-dimensional regime, the leading and trailing electrons generate a well-defined potential that acts as an effective boundary (see the blue curves in Figure 1b), transiently confining the inner electrons during propagation and enabling the same Coulomb-driven ordering mechanism underlying crystallization in trapped and solid-state systems (Figure 1a). This self-induced confinement therefore provides the finite time window required for a freely propagating electron pulse to develop crystalline order. As the beam expands and its aspect ratio decreases, the confining potential gradually weakens and eventually disappears. Nevertheless, if its lifetime exceeds the characteristic ordering time, the electrons can rearrange into a spatially ordered configuration, producing a longitudinal density profile with one distinct peak per electron, a hallmark of a free-electron Coulomb crystal (see the time-dependent densities in Figure 1b.
In electron microscopy, this high-aspect-ratio limit is commonly referred to as the pencil-beam regime, whereas the opposite limit, characterized by , is known as the Holtsmark regime [26]. Both regimes are conventionally described in terms of the statistical broadening of the energy and angular distributions [52], under the assumption that the beam retains a single-peaked phase-space density. This description, however, breaks down when stronger electron-electron correlations generate multi-lobed distributions, as occurs in the present work. In what follows, we therefore adopt a framework based on the melting behavior of a strongly correlated electron chain, combined with a full numerical treatment of the many-body dynamics.
One-dimensional equilibrium model
Because the Coulomb forces between particles in a quasi-one-dimensional arrangement depend primarily on their axial separations and only weakly on their exact transverse positions, the formation of longitudinal order in -beams can be described using a simple one-dimensional model. In this model, the electrons interact along a single line through a softened Coulomb potential within a non-relativistic framework , which we can adopt throughout this work because most of the interaction occurs at kinetic energies below 10 keV.
For a finite number of electrons, a purely repulsive Coulomb interaction cannot form an equilibrium state in the absence of an external restoring potential, which is not present in the setup considered here. Nevertheless, an alternative form of stability can arise by assuming that the required trapping potential takes the form and is generated self-consistently by the two outermost charges, which are treated as fixed (see Figure 1c). This assumption is justified if the ordering of the inner electrons occurs on a shorter time scale than that associated with the evolution of the global pulse geometry. In this regime, the expansion can be treated quasistatically, with the two outermost electrons regarded as effectively fixed while the remaining particles relax toward the equilibrium configuration determined by the instantaneous values of and . Using Monte Carlo simulations (see below), we validate this approximation by comparing the ordering time , with , to the characteristic longitudinal and transverse evolution times, and , respectively. While in the high-aspect-ratio regime the transverse evolution plays only a minor role, we find that, during the high-density stage in which ordering develops, (see Figure A1h).

Figure A1. Evolution of the -beam properties in the absence of lateral trapping. Time evolution of the beam properties as a function of the time elapsed since pulse photoemission, defined from the ejection of the last electron, for different numbers of electrons per pulse (see legend in panel (b)). We show the center-of-mass axial coordinate (a) and energy (b), the linear density (c), the pulse current (d), the transverse beam size (e), the pulse aspect ratio (f), and the longitudinal temperature (see main text) (g). Panel (h) compares the characteristic time scales governing the beam evolution: the ordering time , the transverse expansion time , and the longitudinal expansion time .
In the presence of the self-trapping potential, crystalline order can be quantified through the mean-squared displacement [53] of the particles from their equilibrium positions. We evaluate these fluctuations within the harmonic approximation, following an approach analogous to that used to calculate phonon modes in solids [54]. We first determine the equilibrium positions of the inner electrons by minimizing the potential energy , and then expand it to second order around equilibrium as , where is the corresponding force-constant matrix. For an average kinetic energy characterized by an equilibrium temperature , the positional fluctuations are related to through (see Appendix B).
Mechanical stability is a necessary condition for spatial ordering, whereas thermal fluctuations progressively weaken interparticle correlations. In practice, the absence of mechanical stability leads to imaginary displacements, while thermal motion produces displacements that remain compatible with an ordered chain only below a critical value. This threshold is conventionally determined using the Lindemann melting criterion [55], stating that if the average displacement is larger than a fraction of the average inter-particle distance , the crystal transitions to a gas or liquid phase. The condition leads to the transition-line equation
The curve defined by Eq. (1) depends on only two parameters, the AR and the longitudinal Coulomb coupling , which quantifies the ratio between Coulomb and kinetic energy in the one-dimensional chain model. In Figure 1c, we plot the melting line for by setting over the full range of aspect ratios for which the chain is mechanically stable (AR ). As expected, spatial ordering is favored in electron pulses with high aspect ratios, whereas lower aspect ratios lead to disordered configurations. Stability with respect to stochastic thermal fluctuations is instead captured by the Coulomb coupling which should exceed 0.3 to obtain ordered structures. We emphasize that, because of the long-range nature of the Coulomb interaction, the resulting phase diagram depends strongly on the particle number , making the formation of stable chains increasingly difficult at large particle numbers (see Figure A3c).

Figure A3. Equilibrium positions and normal frequencies of a -electron Coulomb chain. (a) Equilibrium positions normalized to the linear density , as a function of the aspect ratio for chains with different number of particles (see legend in panel c). Each curve terminates at the minimum aspect ratio compatible with mechanical stability. For instance, for , for , for , and for . (b) Minimum normal-mode frequency as a function of the aspect ratio and particle number, normalized to the plasma frequency . (c) Order-disorder transition lines computed from the Lindemann criterion (see Eq. (1)) for a chain containing from to 10 particles and for .
Formation of a transient crystal state in a transmission electron microscope
The time-dependent statistical properties of the photoemitted ensemble of classical electrons are commonly described by the phase-space distribution , which gives the probability of finding the -th particle at position with velocity at a given time. To account for the strong correlations induced by inter-electron scattering, we numerically solve the Liouville equation using a Monte Carlo method, in which the initial particle velocities and positions are sampled from a prescribed distribution and subsequently evolved in time (see Appendix A for more details). In particular, we simulate a photoemission geometry where electrons are emitted from the nanotip and rapidly accelerated by an extractor held at kilovolt potentials before being further accelerated to kinetic energies of several hundred keV by the linear accelerator, as schematically illustrated in Figure 2a.

Figure 2. Coulomb crystal formation in a pulsed electron beam. (a) A laser pulse generates an -electron pulse near a nanotip. The pulse evolves under the influence of accelerating fields provided by the extractor, held at 2 kV, and the accelerator, at 200 kV, located at distances of 500 m and 0.9 m from the tip, respectively. (b) Evolution of the properties of an -beam with during propagation along the column against the phase diagram that determines the conditions for crystallization. We plot the aspect ratio AR and the longitudinal Coulomb coupling parameter , computed using the longitudinal beam temperature and the average linear density (see main text). (I-IV) Longitudinal phase-space density at different -planes corresponding to the specific path coordinates marked in panel (b) (roman numerals). Time and energy marginals are shown in the top and right panels, respectively. (c) Evolution of the -beam properties within the phase diagram for and 8 electrons per pulse. The crystallization region shrinks for larger particle numbers, yielding an exponential decrease in the time the beam path spends within it, as shown by the fit (dashed red line) in the inset. The photoemission parameters used in all panels are reported in Appendix A.
We use these simulations to characterize the confinement experienced by the inner electrons due to the particles occupying the leading and trailing positions of the pulse. A key result is that, despite the random nature of the photoemission process and the resulting shot-to-shot variations in the microscopic configurations, under the studied conditions their ensemble average converges to a well-defined and reproducible axial self-confinement. Remarkably, this averaged potential is accurately described by the one-dimensional softened-Coulomb form constructed from the instantaneous linear density and aspect ratio AR extracted from the Monte Carlo simulations (see Figure A2). The underlying many-particle dynamics can therefore be represented as emerging from a deterministic potential parametrized by the macroscopic beam properties. In this regard, together with the longitudinal temperature (averaged over electrons at a fixed longitudinal plane ), these parameters establish a direct mapping between the evolving e-beam and the corresponding phase diagram.

Figure A2. Effective one-dimensional softened potential. Comparison between the average potential experienced by the inner electrons of the pulse due to the leading and trailing electrons and the effective self-trapping potential , for pulse durations fs (a–d) and fs (e–j). Each column corresponds to a different number of electrons in the pulse, . The average potential is extracted directly from the time-dependent particle simulations according to Eq. (9), whereas is evaluated analytically using the parameters and obtained from the same simulations (see Appendix B).
In Figure 2b we trace, for pulses containing 5 electrons, the path followed by the beam within the phase diagram of Figure 1c. As expected from the rapid initial temporal stretching of the beam, the near-emission configurations yield an average AR and relatively strong Coulomb coupling due to the large pulse current, , which gives a high initial linear density and a moderate initial thermal energy of (see Figure A1). This combination of parameters is clearly located in the crystallization region of the chain phase diagram (see Figure 2b), thus promoting the formation of an ordered state. However, at this stage the electrons still occupy an uncorrelated energy-time phase space, as depicted in Figure 2l. After emission, the beam undergoes several structural transformations: Coulomb scattering leads to a temperature increase and to a three-dimensional beam expansion, thereby reducing the pulse current and the aspect ratio, the latter being further decreased by the initial beam divergence (see Figure A1c–Figure A1g). In terms of the path in the phase diagram, after spending a certain time ordering themselves in the crystal-allowed region, the electrons cross the stability line and remain in the disordered phase until the end of the acceleration stage. Interestingly, close to the crossing of the phase-transition line (Figure 2l), the electrons already appear to form a well-structured time density, where each electron is represented by its own peak, in stark contrast with their single-peaked unstructured energy distribution, which closely resembles an uncorrelated equilibrium distribution. Because of its close correspondence to the spatial densities typical of one-dimensional electron crystals [11] [50], together with the uncorrelated energy distribution, we define this short-lived state of free electrons as a transient Coulomb crystal (TCC).
Once the TCC is formed, the e-beam undergoes a quasi-deterministic expansion characterized by the appearance of multiple separated energy peaks and a better definition of the temporal peaks (Figure 2III). The expansion process, which also cools down the beam and thus yields an increase in (see Figure 2b), is finally completed at very large times, when the number of energy peaks becomes equal to the number of electrons in the pulse and their distribution no longer evolves in time (Figure 2IV).
A completely different behavior would be observed if the TCC state was not reached during the evolution, as displayed by the longitudinal phase-space evolution for higher- pulses (see Figure S1–Figure S1 of the SI [56] for the case). Indeed, even maintaining the same initial emission conditions, increasing leads to several consequences that are detrimental to the formation of spatiotemporal ordering. In this regard, the larger number of charges composing the e-beam leads to a faster three-dimensional expansion, as well as to a modification of the order-disorder phase diagram. Mechanical and thermal stability are weakened for higher and fixed AR, as a consequence of the stronger inner electron repulsion, which can more easily overcome the barrier imposed by -trap (see Figure A3c). In Figure 2c, we show how the area occupied by the crystal region for shrinks in comparison with that occupied for . All these effects reduce the time that the e-beam spends inside the crystallization region, , while ordering itself, displaying an exponential decay as a function of the number of electrons in the pulse (see inset in Figure 2c).

Figure S1. Phase-space evolution for a pulsed beam for different number of particles. (a-h) Longitudinal phase-space density at different -planes for pulses with (a-d) and (e-h). Time and energy marginals are shown in the top and right panels, respectively. Emission parameters are kept the same as in Figure 2 of the main text.
Since the ordering time remains at over the entire initial part of the e-beam dynamics (see Figure A1h), we expect, under the simulated conditions, crystallization for pulses with electrons, whereas selecting higher- electron pulses will lead to a loss of long-range correlations.
Experimental observation of TCC in a transmission electron microscope
To test these predictions, we performed number-resolved spectroscopy of few-electron pulses in the Göttingen Ultrafast Transmission Electron Microscope (UTEM), a platform based on a modified JEOL JEM-F200. Building on our previous studies [31, 37], we access -electron pulses with substantially higher electron numbers. The pulses are generated via near-threshold photoemission from a W(100)/ZrO Schottky emitter, accelerated to 2 keV and 7 keV by an extraction and focusing electrode assembly, and brought to a final beam energy of keV for propagation through the microscope column (see Figure 3a). At the column exit, the electrons are energy-dispersed by an imaging energy filter and recorded by a Timepix3-based event-driven camera with near-unity detection efficiency. The nanosecond-scale temporal resolution of the detector unambiguously correlates each transmitted electron to its laser pulse, giving direct access to the number of electrons () per pulse and its full set of kinetic energies for subsequent spectral analysis. The emitted electron number follows Poisson statistics and scales with the excitation laser power [31, 33]. Unlike previous experiments, we intentionally operate at substantially higher mean electron numbers, up to , accessing events with –9.
Figure 3. Experimental observation of TCC by recorded energy-energy correlations in a transmission electron microscope. (a) Experimental setup. -electron pulses are generated by near-threshold femtosecond photoemission from a Schottky field emitter, accelerated to 200 keV in the transmission electron microscope, energy-dispersed and detected by an event-driven camera, yielding and for each pulse. (b,c) Experimental histograms of the energy difference between the two outermost electrons (b) and all pairs of inner electrons (c) of the energy-sorted pulse, for –9 (dark to light). The outermost pair remains well separated, with a separation growing monotonically with , whereas the energy differences among the inner electrons show a correlation gap at small values that progressively fills in with increasing . (d) Mean energy difference of the outermost (diamonds) and innermost (triangles) adjacent electron pairs versus . Dashed lines: uncorrelated reference from event mixing, i.e., -electron events assembled from electrons of different measured events. The Coulomb-induced excess over this reference persists for the outermost pair but vanishes for the innermost pair with increasing . (e–i) Measured axial energy correlations for (e), 5 (f), 6 (g), 7 (h), and 8 (i) electrons, displaying a transition from a crystalline to non-crystalline beam. (j–n) Monte Carlo simulations reproducing the experimental conditions in (e–i). Maps (e–n) show the joint probability density, normalized to its maximum (linear scale, white to yellow).
A first indication of the internal pulse structure comes from the pairwise energy differences within each event.
In Figure 3b&Figure 3c, we compare the probability densities of the energy difference between the two outermost electrons (b) and between all pairs of inner electrons (c), for –9. The two classes behave in strikingly different ways. The outermost pair exhibits a well-defined, single-peaked distribution centered at a large energy separation that grows monotonically with , reflecting the Coulomb expansion of the leading and trailing electrons. The inner electrons, in contrast, show a correlation gap that progressively fills with increasing . To quantify this, we turn to the mean spacing of adjacent electrons at the edges and in the center of the sorted pulse (Figure 3d). Part of its -dependence is purely statistical: as more electrons share the same spectrum, adjacent energies of the sorted pulse move closer together even without any interaction. We therefore compare the mean spacing of the outermost and innermost adjacent pairs with an uncorrelated reference obtained by event mixing, i.e., from -electron events assembled from electrons of different measured events (Figure 3d, dashed lines). The Coulomb-induced excess over this reference behaves very differently for the two classes. At the pulse edges it persists for all , mirroring the special dynamical role of the two outermost charges, which, per the self-trapping mechanism introduced above, generate the confining potential experienced by the inner particles while remaining only weakly affected by fluctuations in the interior of the pulse. For the innermost pair, the excess gradually vanishes with increasing , signaling that the inner electrons progressively lose their ordering, consistently tent with the predicted disappearance of the crystalline phase at higher particle numbers, while the self-trapping boundary electrons remain well separated throughout.
This picture is confirmed by the complete two-particle spectral statistics. For each electron number, we generate the energy-energy correlation maps in Figure 3e-Figure 3i by binning pairwise energy differences within a pulse into a two-dimensional histogram, with each electron’s energy expressed relative to the pulse’s center-of-mass energy. The maps reveal a pronounced dependence on electron number. For and 5 (Figure 3e&Figure 3f), they break up into a regular lattice of well-separated maxima. This discretized pattern is the spectral fingerprint of the expansion of an initially space-ordered configuration (cf. Figure 2III, IV) and is therefore consistent with the transient formation of a Coulomb crystal during early propagation. With increasing electron number (Figure 3g-Figure 3i), the discrete structure progressively washes out from the interior outward: the inner-electron maxima broaden and merge, while the outermost features persist, in agreement with the pair-resolved analysis of Figure 3b-Figure 3d. For , the map approaches the smooth distribution expected for a stochastically expanding, disordered ensemble.
These observations are quantitatively reproduced by Monte Carlo simulations carried out under the experimental emission and acceleration conditions (Figure 3j-Figure 3n). The agreement between measured and simulated maps, including both the persistence of the outer-pair anticorrelations and the gradual loss of inner-electron discretization with increasing , supports the interpretation that the final-state energy ordering is inherited from a transient, spatially ordered configuration formed during initial propagation, rather than emerging only during the final acceleration stage.
Crystallization of trapped N-electron pulses
Since the formation of spatially ordered structures is promoted by preserving a large beam aspect ratio over extended propagation times, a natural strategy to mitigate the loss of long-range order at high is to axially constrain the electron trajectories, i.e., to refocus the electrons. Furthermore, this must be done in such a way that Coulomb collisions do not lead to excessive heating of the beam, thereby pushing it toward small values of . Lateral trapping of charged particles can be achieved in several ways, either by resorting to static electrodes or to electrodynamic potentials. In this work, we propose two practical designs to achieve lateral confinement, namely by ponderomotive forces shaped in a Laguerre-Gauss mode and a series of static quadrupoles with alternating polarity. Both mechanisms are described in more detail in Appendix C.
Electron-electron correlations of trapped few-electron pulses
Because, in both trapping mechanisms, the effective force can be approximated as harmonic, we include this force during an acceleration-free stage of length placed between the end of the extractor and the linear accelerator (see Figure 4a), while maintaining the same configuration as in Figure 2. We investigate the dependence of the energy correlations on the trapping length and the number of particles per pulse at a fixed, relatively low trapping frequency of (see Table I for a proposed set of experimentally feasible parameters that yield such value).

Figure 4. Trapping of -electron pulses. (a) A photoemitted electron pulse composed of electrons is accelerated by the extractor and then injected into a harmonic transverse trap with trapping frequency , realized either by sending the beam through a Laguerre-Gauss (see inset) mode via ponderomotive forces or by a series of static quadrupoles with alternating polarity. After the trapping stage of length , a pulsed Coulomb crystal beam is formed thanks to the one-dimensional confinement provided by the trap, and is then brought to keV by the accelerator. (b,c) Beam aspect ratio (b) and average transverse electron-electron distance (c) as a function of time elapsed since the photoemission of the pulse, for electrons and for different trap lengths. (d-o) Energy-energy correlations after the final acceleration stage for (columns) electrons per pulse, with no trap (uppermost row) and with and mm (middle and bottom rows). A trapping frequency of , achievable with state-of-the-art laser systems and miniaturized electronics (see Appendix C), is used in all panels. Initial source parameters are kept the same as in Figure 2.
| Trapping frequency | Ponderomotive guiding NA | Ponderomotive guiding | Ponderomotive guiding | Ponderomotive guiding | Ponderomotive guiding | Electrostatic quadrupolar guiding | Electrostatic quadrupolar guiding | Electrostatic quadrupolar guiding | Electrostatic quadrupolar guiding | Electrostatic quadrupolar guiding | Electrostatic quadrupolar guiding |
| 10 | 0.0041 | 82 | 1.17 | 2.0 | 262 | 100 | 0.77 | 141 | 2 | 0.2 | |
| 10 | 0.0036 | 70 | 1.55 | 1.9 | 175 | 100 | 1.16 | 94.3 | 2 | 0.3 | |
| 10 | 0.0029 | 58 | 2.33 | 2.0 | 33 | 50 | 1.55 | 70.7 | 2 | 0.4 |
Table I. Ponderomotive and quadrupolar trapping. Different choices of parameters yielding a harmonic trapping frequency . We report parameters for ponderomotive guiding with a Laguerre–Gaussian laser pulse of electric-field standard deviation with and , for which , pulse energy , and waist . The optical parameters are chosen such that the Rayleigh length is and for a pulse duration of . For electrostatic quadrupolar guiding, the parameters are chosen for 2 keV electrons and lie within the stability region ; the trapping frequency is computed as , with , where is the electron velocity and is the quadrupole pitch.
The effect of the trapping stage can be followed by examining the beam dimensions as a function of the time elapsed since the emission of the last electron. Immediately after the extractor, the beam is refocused every half period (), until it exits the trapping stage after a time ( for ), before being accelerated again to 200 keV (see Figure 4b,Figure 4c). Under this sequence of focusing and refocusing events, Coulomb repulsion induces an axial expansion of the beam, bringing the aspect ratio back to very high values at the expense of a rapid decrease in the linear density, as shown in Figure 4b. For instance at , this rapid expansion, together with moderate beam heating, brings the Coulomb coupling to thus right below the boundary defining the crystal region of the phase diagram placed at (see Figure A3c) even if in a high-aspect-ratio regime. Nevertheless, this does not imply a sudden loss of correlations: in the vicinity of the crystallization boundary, the system still retains a strongly correlated liquid character. This can be appreciated by considering the Lindemann condition in Eq. (1) which, for high aspect ratio and allowing fluctuations up to three times the average lattice spacing, i.e., , places the corresponding order-disorder line at (). As a result, the electrons complete the propagation with very strong energy-energy and time-time correlations. Similar arguments apply to the other explored electron numbers.
We notice that even a single refocusing of the beam () enhances electron-electron correlations in the final accelerated beam with respect to the trap-free case. Indeed, comparing panels d–g with h–k in Figure 4 reveals a marked difference, especially for small numbers of electrons. Remarkably, in this configuration, the cross-over promotes ordering within the -electron rather than solely producing the detrimental statistical broadening typically associated with cross-overs in electron microscopy [57]. By contrast, a longer trapping stage is required for pulses containing a larger number of particles. As shown in Figure 4l–Figure 4o, trapping over at 2 keV is sufficient to produce crystalline pulses containing up to electrons per pulse.
Continuous beam crystallization
The preceding results demonstrate that transverse guiding can extend the timescale over which Coulomb ordering develops in few-electron pulses and preserve strong correlations at higher particle numbers. However, the finite longitudinal extent of these pulses ultimately limits the persistence of their periodic structures, as Coulomb repulsion drives longitudinal expansion. This limitation motivates the investigation of ordering in continuous electron beams, where the number of electrons approaches infinity and longitudinal expansion is suppressed. In this regime, Coulomb interactions may give rise to a stationary, strongly correlated states exhibiting spontaneous temporal bunching, opening a potential route to generating structured electron beams at high average currents.
Here, we investigate this regime for transversely guided electron beams in electron microscopes. Related systems have been studied in the context of ion Coulomb crystals in storage rings [3, 4, 5], where beam shearing can disrupt spatial ordering [58]. In contrast, electrons in electron microscopes predominantly propagate along straight trajectories. We therefore focus on the equilibrium spatial properties of trapped continuous electron beams and identify the physical conditions under which Coulomb-induced ordering forms.
Equilibrium properties of a trapped continuous Coulomb-bunched beam
We consider an initially uncorrelated beam entering the trapping stage and subsequently thermalizing within the confining potential to form an equilibrium state (see Figure 5). After a sufficiently long thermalization time [59, 60], the phase-space distribution approaches the Boltzmann distribution (see Eq. (12)).

Figure 5. Crystallization of a continuous electron beam. (a) A continuous source produces an electron current , which is accelerated by the extractor and injected into a transverse trap with frequency , similarly to Figure 4a, but with a trap length sufficient for thermalization. The thermalized beam exits in a gas, liquid, or crystal phase depending on the Coulomb coupling , with the Wigner-Seitz radius , and on the normalized linear density . (b-d) Longitudinal phase-space density at the trap exit for (b), 0.2 (c), and 0.7 (d), at fixed . As decreases, the beam undergoes crystal-liquid and liquid-gas transitions, as shown by the density-density correlation functions in the insets. Here, meV and keV. (e-g) Beam density cross-section as a function of . For higher-order structural phases relative to the string configuration, the beam evolves from a single-peak distribution to a donut-shaped profile. We set as in b-d. (h) Universal plot (see main text) of the duty cycle of the Coulomb-bunched beam as a function of and . The 50% contour line highlights the transition between disordered and ordered phases, while the horizontal dashed line at marks the string-to-zig-zag structural transition. (i) Coulomb crystal beam impinging on a sample and yielding cathodoluminescence (CL) with a superradiant photon-emission rate scaling as . (j) Superradiant CL enhancement factor for a Coulomb beam as a function of at .
Because of the characteristic spatial scaling of the Coulomb interaction, a one-dimensional lattice cannot sustain true crystalline order at any finite temperature. As a result the Lindemann-criterion approach employed above for finite is not suitable for characterizing particle thermal fluctuations (see SI [56]). We therefore employ standard molecular-dynamics methods to exactly sample (see Appendix § 0.9.4). Scaled beam properties are universally governed by two dimensionless parameters: the normalized linear density and the equilibrium Coulomb coupling . Here, is the Wigner-Seitz radius, which is set directly by the transverse trapping frequency.
At fixed trapping frequency, beam ordering can be achieved either by decreasing the average distance between electrons or through cooling. By increasing the linear density from to , the beam transitions from a gas phase, showing a spatially flat longitudinal phase-space density, first to a liquid and then to a strongly ordered structure that resembles a solid but still shows a decay in the density-density correlation function (see Figure 5b-Figure 5d and the corresponding insets for a scan along the line). For values above the critical value , the internal Coulomb repulsion pushes the electrons away from the minimum of the trapping potential, yielding a structural transition from a one-dimensional string to higher-order structures such as zig-zag and helical configurations [61]. This beam conformation can be directly visualized in the transverse density, which shows a single peak in the string phase (Figure 5e,Figure 5f) and a donut-shaped profile in the higher-order phases (Figure 5g).
As the beam transitions further into the crystalline state, the traveling electrons occupy increasingly narrow temporal regions, whose duration is limited by fluctuations determined by the beam temperature, linear density, and trapping frequency. For this reason, the beam duty cycle , i.e., the ratio between the time duration of one pulse and the crystal period (see sketch in Figure 5h), provides a convenient measure of crystallization and of the degree of longitudinal compression. We define , where is the effective temporal duration of a single-electron pulse obtained from a generalized definition of the longitudinal beam emittance (see Eq. (15)). A duty cycle of 50% provides a useful threshold separating the disordered and crystalline regimes (see Figure 5h). Interestingly, even in the limit of high Coulomb coupling, , and large normalized density , the duty cycle approaches the non-zero value of due to interaction-induced fluctuations.
Superradiant light emission from continuous Coulomb crystals
Longitudinal electron bunching plays a central role in the generation of coherent radiation by free-electron beams. In relativistic free-electron lasers, microbunching can develop self-consistently through the collective interaction between electrons and the radiation field, as in self-amplified spontaneous emission, or be externally induced through coherent seeding schemes [45]. At lower electron energies, laser-induced bunching enables coherent cathodoluminescence (CL) through coupling to nanostructures [62, 63], coherent control over the excitation of quantum systems [37], laser-electron pump-probe experiments [64], and superradiant scaling of light emission [65–69]. However, in these low-energy approaches, temporal bunching does not imply correlations between individual electrons, which remain uncorrelated. In contrast, the longitudinal ordering considered here emerges spontaneously from bare Coulomb interactions between electrons, without externally imposed temporal modulation. We therefore compute the probability associated with superradiant CL emission from such self-organized beams, both as a potential route to coherent light emission and as a direct optical probe of the degree of beam crystallization.
For the strongly correlated electrons described by the equilibrium distribution , the angle-integrated CL emission rate accumulated over an illumination time grows linearly with time, with a proportionality factor that scales quadratically with the electron-beam current. Here, denotes the single-electron CL emission probability per unit frequency [21], as derived in Appendix G. The quadratic current dependence is a signature of superradiance, whose magnitude is governed by the enhancement factor
Because the bunching is determined solely by Coulomb repulsion, the temporal spacing between electrons is directly set by the beam current. We therefore expect coherent emission from a crystalline beam to appear at frequencies , with a positive integer. By taking the Fourier transform of the density-density correlation function (insets in Figure 5b-Figure 5d), we compute Eq. (2) as a function of frequency and Coulomb coupling (see Figure 5j). Interestingly, this quantity clearly identifies the transition from a disordered to a crystalline structure through the appearance of multiple orders of coherent emission as the Coulomb coupling increases. In this regard, the enhancement factor shows a strong coherent build-up of radiation already for , reaching values at the first harmonic and considerably smaller values for the second and the third (Figure 5j).
Requirements on the input beam to achieve continuous beam crystallization
While the preceding analysis establishes the beam properties attainable under certain equilibrium conditions, assessing whether such states can be reached requires relating them to the characteristics of the beam entering the trap: its transverse size , divergence or convergence semi-angle , and energy spread . Since the beam current is conserved, the equilibrium temperature is the only remaining quantity to be determined from these initial conditions. To this end, we impose conservation of the average energy per particle between the incoming beam and the equilibrium state characterized by . Assuming that the Coulomb energy of the incoming beam equals that of the zero-temperature equilibrium configuration, corresponding to perfect axial ordering, we derive an upper bound on the equilibrium coupling parameter and determine the trapping frequency that maximizes it. This procedure yields the optimal frequency , together with the corresponding bound on the equilibrium Coulomb coupling (see Appendix F for a detailed derivation)
Interestingly, according to , beams with larger require weaker transverse confinement, because their larger initial potential energy is converted into thermal energy as the system approaches equilibrium.
Eq. (A15) predicts a considerable dependence on longitudinal and transverse beam temperatures, i.e., on the center-of-mass energy, energy spread, and convergence angle. In particular, low-energy beams are favored for more monochromatic beams, whereas higher energies should be chosen for beams with large energy spread. For instance, we obtain with µeV for a beam with eV, focused into the trap with nm and convergence angle mrad at eV (see Figure A4f-Figure A4i). This would form a beam with a good level of crystallization, corresponding to a duty cycle of if formed at a current µA (see Figure 5h). High transverse beam temperatures strongly undermines crystallization. Indeed, a beam with equal divergence and lateral size but accelerated at 2 keV and with current µA would not crystallize, since even with a higher required trapping frequency of µeV (see Figure A4f-Figure A4i).

Figure A4. Relation between optimal input and equilibrium beam properties (a) An input continuous electron beam with current , transverse size , divergence half-angle , and energy spread enters a trap with an optimal frequency at a kinetic energy . After thermalization, the beam exits the trap in a state characterized by a duty cycle and a temporal spread . (b-e) Optimal trapping frequency (b), input current (c), and output pulse temporal duration (d) and duty cycle (e) as a function of and for and . (f-i) Same as in panels b-e but for . In all panels to reach the minimum duty cycle with highest temperatures (see Figure 5h).
These results highlight that continuous-beam crystallization is not determined by the electron current alone, but requires a suitable balance between the initial beam properties and transverse guiding. In particular, high linear density must be combined with sufficiently low longitudinal and transverse energy spreads to prevent thermal
fluctuations from suppressing Coulomb-induced ordering. The resulting conditions arising from Eq. (A15) connect the equilibrium phase diagram in Figure 5h to experimentally accessible beam parameters, providing a practical guideline for the generation of crystallized continuous e-beams.
Conclusions
Ordered forms of matter are ubiquitous in nature and range from solid-state systems, where atoms attract each other, to charged-particle systems, where mutually repelling electrons or ions can form Coulomb crystals when confined by an external field. In this work, we have shown that the paradigm connecting Coulomb crystallization with external confinement breaks down in photoemitted few-particle electron pulses. Under suitable low-dimensional conditions, the ensemble-averaged Coulomb potential created by the two outermost electrons gives rise to a well-defined effective potential that acts as a dynamical boundary for the inner electrons. As a result, the beam can self-trap for a sufficiently long time for the electrons to form an ordered state (Figure 1). Remarkably, since this ordering develops on a faster timescale than beam expansion, the conditions for crystallization at finite electron number can be predicted through a phase diagram defined solely by macroscopic dynamical beam parameters. During propagation, this mechanism drives the dynamical evolution of the beam into what we call a transient Coulomb crystal (TCC), in which the electrons are strongly correlated in time but remain uncorrelated in energy (Figure 2).
While this behavior was demonstrated through particle simulations, we also provided experimental evidence of the final product of the expanded TCC by recording energy maps of – electron pulses in a transmission electron microscope. In the experiment, we observed signatures of TCC formation for pulses containing 4, 5, and 6 electrons (Figure 3), whereas higher- pulses remained uncorrelated because the stronger intrabeam Coulomb expansion, together with the increased mechanical and thermal instability at larger particle number, prevents the beam from ordering within the time spent in order-favored region of the phase diagram.
We then explored how to sustain the pulse self-trapping potential for longer times by introducing a eV transverse trap during beam propagation in the microscope. Remarkably, we found that only a few refocusing events, corresponding to a 20 mm-long trap, are sufficient to synthesize electron-beam Coulomb crystals with tens of particles (Figure 4). We provided values to implement two viable guiding designs: using free-space ponderomotive forces with a intensity profile displaying a minimum in the center, or stacking a series of quadrupoles with alternating polarity along the direction of electron motion. The former could be implemented using commercially available spatial-light modulators [70–72] and by exploiting recent advances in light-induced free-electron shaping [73, 74]. The latter could benefit from rapid progress in micro- and nanofabricated charged-particle optics [75–77], which can provide electric fields of order V/m over distances of tens of microns.
As a further step, we explored the possibility of creating Coulomb crystals from a continuous electron source (Figure 5). Because fluctuations increase with the number of particles, continuous crystallization imposes severe constraints on the input beam entering the trap, requiring simultaneously high current, small transverse size, low divergence, and narrow energy spread. Nevertheless, recent progress in electron-source technology suggests that these requirements are not so out of reach. Cold-field emitters and laser-driven nanotips already provide high brightness, high coherence, and sub-eV energy spreads [51, 78, 79], while nanostructured LaB field emitters have demonstrated ultrabright monochromatic emission and stable microampere-level operation [80, 81]. In parallel, ultracold electron sources based on near-threshold photoionization of laser-cooled atoms offer a complementary route toward extremely low transverse temperatures and low emittance, although so far mainly in pulsed operation [82, 83]. In this regard, a remaining fundamental challenge is the heating induced by stochastic intrabeam scattering at high current [84]. In crystalline ion beams, this problem can be counteracted by laser cooling [6]; however, an analogous strategy can-
not be directly transferred to e-beams because free electrons lack internal transitions. This highlights the need for applying to e-beams cooling or phase-space control methods capable of reducing the three-dimensional effective electron temperature without sacrificing a significant fraction of the beam current, such as stochastic [85], cavity-assisted sideband [86] and Doppler cooling [87].
Beyond the fundamental interest associated with the formation of this novel type of Coulomb crystal, such strong correlations could be harnessed to address several challenges in electron microscopy. Marked temporal and energy correlations between electrons could be exploited to extract electron bunches with pre-selected properties. For instance, electron-current fluctuations could be strongly suppressed by applying filters that select a precise energy distribution corresponding to a definite electron number [31, 39]. Time-ordered electron distributions, tunable either through the degree of lateral confinement or by placing the sample at different axial planes, could also drive superradiant THz sources based on CL emission with tunable quantum statistics [69], and enable new forms of electron microscopy based on timed multi-electron pulses.
More generally, the combination of Coulomb repulsion with external static and time-dependent fields creates a rich many-body dynamics that can open new routes not only for controlling free-electron beams, but also for studying strongly correlated systems in free space. This setting offers direct access to diagnostic tools while allowing the surrounding electromagnetic environment to be modified with a high degree of flexibility. This idea could be further extended by tailoring the electron-electron interaction through nearby dielectric or nanophotonic structures, opening the way to a new research direction devoted to strongly correlated beams with tunable inter-particle interaction.
Acknowledgements
We acknowledge the continued support from the Göttingen UTEM team. We thank Martin Kozák for insightful discussions. The experiments were conducted at the Göttingen UTEM Lab, funded by the Deutsche Forschungsgemeinschaft (German Research Foundation) through Project-ID 4326830600 (SFB 1456, project C01 to C.R.), the Friedrich Wilhelm Leibniz program (RO 3936/4-1 to C.R.) and the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 101017720 (FET-Proactive EBEAM).
Author contributions
C.R. directed the study. V.D.G. prepared the figures, with contributions from R.H.. V.D.G. developed the theoretical framework and conducted all particle and molecular dynamics simulations. R.H. conducted the experiments and data analysis. All authors discussed the results and interpretation. V.D.G., R.H., and C.R. wrote the manuscript.
Competing Interests
The authors declare no competing interests.
Appendix
Particle simulations of photoemitted -electron pulses
We compute the evolution of the phase-space density related to the photoemitted electrons using Monte Carlo trajectory simulations. For each realization, the initial positions and velocities are sampled from the photoemission distribution (see below), and the classical equations of motion are integrated,
where and with . Ensemble averages over the resulting trajectories yield the time-dependent phase-space density and the beam observables discussed in the main text.
Photoemission from the nanotip is modeled following Refs. [31, 37]. Electrons are emitted uniformly from a flat circular surface of radius 100 nm. Their emission times and initial kinetic energies are sampled from Gaussian distributions. The emission-time distribution has zero mean and standard deviation , with fs, while the kinetic-energy distribution has mean eV and standard deviation eV. The initial emission angle is restricted to mrad, as determined by comparison with experimental results.
The simulated propagation consists of an extraction stage, an optional transverse guide of length , and final acceleration to 200 keV. The external force is modeled piecewise in these three regions. In the extraction region , with m and kV, we use the sphere–plane approximation [88] and retain its axial component with m. Within the transverse guide , the force is harmonic, , where eV (see Appendix C). Finally, in the acceleration region , with cm and kV, the electric field is approximated as uniform, . Figure A1 shows the simulated evolution of the beam properties in the absence of transverse confinement .
Fluctuations of a finite electron chain: softened-Coulomb model
To describe the stability of a freely expanding -beam, we consider an effective one-dimensional model in which the electrons occupy a segment of length , with linear density , and interact through the softened Coulomb potential , where is the characteristic transverse electron-electron separation extracted from the particle simulations.
In this model, the leading and trailing electrons are fixed at , whereas the remaining electrons are allowed to relax and fluctuate around their equilibrium positions. The two boundary electrons generate an effective self-trapping potential, , with . The effective Hamiltonian of the inner electrons can then be written as , where
We determine the equilibrium positions by minimizing the potential energy in Eq. (5). Small fluctuations around this configuration are governed by the force-constant matrix
with . Here, , whereas the sum over also includes the two fixed boundary electrons.
Expanding the potential energy to second order around the equilibrium positions yields the harmonic approximation , up to an additive constant, where . Since is real and symmetric, it can be diagonalized in terms of orthonormal normal modes satisfying
Mechanical stability requires all eigenvalues of to be positive, or equivalently, all normal-mode frequencies to be real. The equilibrium positions and lowest normal-mode frequencies are shown in Figure A3a,Figure A3b.
To analyze the thermal stability of the chain, we express the displacement vector as a linear combination of normal modes, . In this basis, the harmonic potential energy becomes . Averaging over the corresponding Boltzmann distribution then gives . Consequently, the mean-square displacement of the -th inner electron is
where the last equality follows from the orthonormality of the eigenvectors of . Averaging over the inner electrons yields . We determine the order-disorder boundary using the Lindemann criterion , with for the phase diagrams in Figure 1,Figure 2, and Figure A3. The resulting melting lines for different are shown in Figure A3c.
The model just described relies on two main assumptions: (i) the validity of the softened Coulomb potential as an effective self-trapping mechanism; and (ii) the separation of the time scales associated with the ordering of the inner electrons and the transverse and longitudinal expansion of the -pulse. To validate assumption (i), we compare with the potential energy experienced by the inner electrons in the trajectory simulations. Denoting the leading and trailing electrons by , with coordinates , the simulation-based potential is given by
At each propagation time, the parameters of the effective model are determined from the simulated pulse length, through , and the corresponding mean transverse electron-electron separation . As shown in Figure A2, the potential in Eq. (9) is accurately reproduced by . This agreement establishes a direct connection between the stochastic many-particle dynamics and the effective description in terms of macroscopic beam parameters.
Assumption is justified by comparing the characteristic ordering time , where , with the longitudinal and transverse expansion times, and , respectively. As shown in Figure A1h, during the high-density stage in which ordering develops. Although can become comparable to , the large beam aspect ratio makes the effective interaction only weakly sensitive to variations in .
Lateral trapping designs
We consider two experimentally feasible strategies for realizing the weak transverse harmonic confinement used in the particle simulations shown in Figure 4: an alternating sequence of electrostatic quadrupoles and ponderomotive guiding by a structured optical field (see Figure 4a). Both schemes generate an approximately harmonic potential in the transverse plane. In particular, alternating static quadrupoles produce an effective Paul-trap potential [89, 90] in the beam frame, whereas the dark central core of a Laguerre-Gaussian optical mode [91, 92] provides confinement through the ponderomotive force.
For electrostatic quadrupolar guiding, the spatially periodic field is experienced as a temporal modulation by electrons traveling with mean velocity . This situation is analogous to the Paul-trap pseudopotential and has been implemented for electron guiding [93]. In this configuration, the electron beam samples the modulation at , with stability parameter , where denotes the applied quadrupole-voltage scale, the effective distance from the beam axis to the electrodes, and the pitch of the alternating sequence. The resulting secular trapping frequency is (see SI [56])
Ponderomotive guiding exploits the cycle-averaged potential associated with a free-space electric field , which drives electrons toward minima of the optical intensity. We use a Gaussian laser pulse of duration , prepared in the mode and with peak electric-field-envelope amplitude . The beam is focused by a lens with numerical aperture and has a Rayleigh range . In this configuration, the harmonic trapping frequency generated near the on-axis intensity minimum at the focus is (see SI [56])
Table I reports representative quadrupolar and ponderomotive parameter sets yielding . For the quadrupolar implementation, the parameters are chosen in the conservative range , which provides appreciable confinement while limiting the micromotion. In the ponderomotive case, to obtain confinement that remains nearly constant over the trapping length required in the particle simulations of Figure 4, we choose a Rayleigh range exceeding .
Simulations of equilibrium properties of a trapped continuous electron beam
To characterize a continuous electron beam beyond the harmonic approximation (see SI [56]), we perform Metropolis simulations [94, 95]. Specifically, we consider a beam composed by electrons evolving in a squared box of size with periodic boundary conditions and subjected to a transverse harmonic potential with trapping frequency . The Hamiltonian governing the stable configuration of the system is given by
To accurately account for the long-range Coulomb interaction while minimizing finite-size and surface effects, we employ the Ewald summation method for charges embedded in a uniform positive background, corresponding to the jellium model [7] [61, 94]. Within this framework, the Coulomb pair-interaction energy is replaced in Eq. (12) by the Ewald potential energy
where the summation runs over the lattice vectors of a squared lattice, and denotes the complementary error function [96].
We implement the Metropolis algorithm by constructing a Markov chain in which, at each step, one of the electrons is selected at random and subjected to a trial displacement. This procedure samples the Boltzmann distribution . To visualize and characterize the Coulomb-induced spatial ordering, we calculate the density and the density-density correlation function along the periodic direction [97].
An important consequence of the structure of is that any quantity of the form depends only on and , rather than separately on . Consequently, the average energy per particle normalized to
where , is a function only of the two parameters and . We exploit this scaling property to determine both the duty cycle in Appendix E and the final equilibrium temperature in Appendix F for arbitrary trapping frequencies while performing simulations at only a single value of .
Calculation of longitudinal emittance and duty cycle
The root-mean-square (rms) emittance is the conventional quantity used to characterize the phase-space area occupied by an electron beam. For a one-particle phase-space distribution , where and , it is defined as , where and . For a single Gaussian distribution, provides a direct and robust measure of the occupied phase-space area.
In the strongly correlated regime considered here, however, the longitudinal phase-space distribution can develop several separated lobes. In this case, the rms emittance need not faithfully represent the physically occupied area. We therefore define a generalized emittance from the union of the contours satisfying .
The generalized emittance is the enclosed area divided by ,
This definition is independent of the detailed distribution shape and reduces to for a single Gaussian lobe. It is therefore suited to the multi-lobed distributions obtained in the equilibrium simulations of Appendix § 0.9.4.
At equilibrium, the single-particle distribution assumes the separable form which transforms Eq. (14) into
where denotes the intervals over which . For a Gaussian spatial profile, , this expression reduces to . Since the length occupied by the electrons in the velocity part of the phase space is given by , the spatial segment is given by .
For the longitudinal direction, , we use the same construction to define the duty cycle as , i.e., the average time occupied by an electron peak respect to the bunching period , with
Here, is the equilibrium single-particle longitudinal density and is the interval associated with the -th density peak. In the string regime, where the distribution consists of identical peaks with linear density , the duty cycle becomes . If each peak is approximately Gaussian with spatial standard deviation , the duty cycle reduces to , where .
Conditions for continuous electron beam crystallization: equilibrium temperature through average conservation of energy
In this section, we determine the beam characteristics required at the entrance of the trap to reach a given equilibrium phase, assuming that the trapping structure is sufficiently long for the beam to thermalize [59, 60]. Because the Hamiltonian is time independent, the average energy is conserved, i.e., . We denote by the mean energy of the beam when it is fully contained within the trap at time , and by its energy after equilibration. Imposing and dividing by the number of particles yields
Since the current and the center-of-mass velocity are conserved during propagation through the acceleration-free trapping region, the only remaining unknown is , which can be obtained by inverting Eq. (16). Together with and , this quantity determines the equilibrium phase reached by the beam for a given input configuration. In principle, all terms on the right-hand side of Eq. (16) should be obtained from nonequilibrium particle simulations, whereas is calculated using the Metropolis algorithm described in Appendix § 0.9.4. Figure S2 in the SI [56] shows the equilibrium energy per particle over the relevant parameter range.

Figure S2. Energy per particle for a trapped continuous beam (a) Universal plot relative to energy per particle computed from the equilibrium simulations of a continuous trapped beam as a function of the Coulomb coupling and the normalized density . Simulations are performed with but values are invariant under changes of the trapping frequency. (b) Cuts along from panel (a) of the energy per particle for several values. (c) Dimensionless coefficients as a function of of a quadratic fit of the normalized energy per particle in panel b.
Because depends monotonically on , Eq. (16) can be inverted to determine the equilibrium Coulomb coupling. The equilibrium energy per particle is accurately described by a quadratic function of with -dependent coefficients (see the scaling properties discussed in Appendix § 0.9.4), . The only positive root is
where must exceed the zero-temperature equilibrium energy , corresponding to , and where all coefficients remain positive over the range of considered here.
To estimate without nonequilibrium simulations, we approximate the input Coulomb-energy contribution in by the average potential energy of the equilibrated beam at zero temperature and fixed . This amounts to setting in Eq. (17). The remaining energy contributions can be expressed in terms of experimentally relevant beam parameters through , , and , where and denote the input angular and transverse beam spreads, respectively. We then obtain
with the residual beam energy and the center-of-mass energy.
Numerical simulations show that the quadratic coefficient is negligible compared with over the entire range of explored in this work (see Figure S2 in the SI [56]). Equation (18) can therefore be approximated as , which reaches a maximum at the optimal trapping frequency
This frequency yields the maximum Coulomb coupling
Because is obtained by assigning the minimum Coulomb energy to the input beam at fixed , any other input beam with the equal properties but with a different particle spatial distribution satisfies .
In Figure A4, we show the effect of the input-beam properties on the equilibrated bunch duration and duty cycle for (Figure A4b–Figure A4e) and (Figure A4f–Figure A4i) electrons. In all cases, we use , which approximately provides the smallest duty cycle at small values of (see Figure 5h) and yields an optimum current .
CL emission from a strongly-correlated continuous e-beam
We now derive the CL emission probability generated by the interaction of correlated electrons with a sample capable of supporting coherent CL emission [21]. We begin with the angle- and frequency-resolved, time-integrated far-field photon-emission probability produced by the fluctuating current density associated with electrons propagating with mean velocity . This probability is given by [64]
where , , and the electromagnetic fields are expressed in terms of the Green tensor as and [21]. For a dielectric environment with a linear response described by a spatially and frequency-dependent permittivity , the Green tensor is implicitly defined by the equation . The average in Eq. (19) is taken over the Boltzmann distribution associated with the Hamiltonian .
Because the interaction between electrons and optical excitations depends only weakly on over electronvolt variations in kinetic energy, we neglect fluctuations of this quantity within the beam. Moreover, the characteristic transverse size of a beam in the string phase, ( and ), can be reduced to few nanometers using electron optics. We therefore assume that the CL emission varies negligibly across the transverse extent of the crystalline beam and set . Under these assumptions, Eq. (19) can be recast as
where denotes the single-electron CL emission probability density. It is defined by Eq. (19), evaluated using the current density .
For a continuous beam, the second term in Eq. (20) can be expressed in terms of the density–density correlation function as
where is the superradiant enhancement factor introduced in the main text. Importantly, in the limit of at fixed , the enhancement factor remains finite and in particular it is bounded by . Finally, integrating Eq. (19) over all solid angles and dividing by the measurement time yields the total photon-emission rate per unit frequency, . This result exhibits a quadratic dependence on the current , together with a coherent contribution that builds up linearly with the measurement time.
Coulomb Crystallization in Free Electron Beams – Supplementary Information –
Valerio Di Giulio; Rudolf Haindl; Armin Feist; Claus Ropers
Quadrupole and ponderomotive lateral trapping designs
In this section, we derive the effective transverse forces produced by two guiding configurations: an alternating sequence of static electrostatic quadrupoles and the ponderomotive force generated by a structured optical field. We consider a beam whose mean propagation direction is the axis and whose mean longitudinal velocity is .
Guiding with an alternating sequence of static quadrupoles
Consider a sequence of static quadrupoles arranged along the direction at a transverse distance from the central axis. Within each element, two opposing electrodes are held at a positive potential, while the other two are held at
a negative potential. The polarity is reversed between consecutive elements, so that a transverse direction that is focusing in one element becomes defocusing in the next. Near the symmetry axis, the electrostatic potential of an ideal quadrupole can be written as , where denotes the magnitude of the quadrupole gradient and is a dimensionless function describing the alternating polarity. For an idealized periodic sequence, where is the spatial period, or pitch, of the structure. The precise relationship between , the applied electrode voltage , and the electrode separation depends on the electrode geometry and the fringing fields. We incorporate this geometrical dependence into the effective definition of and write , where is a dimensionless geometrical factor that can be determined analytically or numerically for a specific trap design [1] and is absorbed here into the definition of . Neglecting the force component along the electron trajectory, the transverse force acting on an electron is . At any fixed longitudinal position, the field therefore focuses the electron along one transverse direction while defocusing it along the orthogonal one.
For a particle moving approximately uniformly along , the spatially periodic field is sampled as the time-dependent modulation , whose fundamental temporal frequency is consequently . For a sequence of discrete quadrupoles, the modulation can therefore be expanded as , yielding the transverse equation of motion
with the matrix and the stability parameter . We separate the trajectory into a slowly varying secular component and a rapidly oscillating micromotion coordinate , with . Over a single modulation period, can be regarded as approximately constant. To first order in , the rapidly oscillating component satisfies . We restrict the following discussion to the simple case of a cosine modulation, for which . A more general treatment of stability and transverse trapping can be formulated using the transfer matrix approach, but lies beyond the scope of this work. The rapidly oscillating component is then . Substituting into the equation of motion and averaging over one modulation period gives
Thus, the time-averaged transverse dynamics are governed by the harmonic equation , with trapping frequency , while the approximate full trajectory is . For a cosine modulation, the stability range follows from the first Mathieu stability region, which extends from to [2, 3]. Operating near the upper stability boundary is generally undesirable because the micromotion becomes large, the pseudopotential approximation loses accuracy, and the trajectory becomes increasingly sensitive to field imperfections and higher spatial harmonics. The more conservative interval therefore provides appreciable confinement while limiting the micromotion amplitude to approximately of the secular displacement. In Tab. 1 in the main text, we report three practical implementations of this trapping scheme that yield .
Guiding by free-space optical ponderomotive forces
Ponderomotive forces have been employed for both elastic and inelastic electron-beam shaping and can also provide transverse guiding [4, 5, 6, 7]. We consider an electron interacting with a monochromatic laser field of frequency propagating along the direction. The vector potential is written as , where is a slowly varying complex envelope. Starting from the minimal-coupling Hamiltonian and approximating the electron momentum as , we obtain, after omitting the field-independent kinetic-energy term,
We next average Eq. (21) over the rapid laser oscillations, treating the envelope as constant over one optical cycle and approximating the electron trajectory as . Under this averaging, the term linear in the vector potential vanishes, whereas the quadratic term yields the ponderomotive potential energy
This potential drives the electron toward regions of lower laser intensity. The corresponding transverse guiding-center dynamics is governed by .
A Laguerre–Gaussian mode provides the spatial intensity profile required for transverse confinement. For a pulsed laser focused at the axial position by a lens of numerical aperture NA, the electric field is [8]
where and are the azimuthal and radial indices, respectively, is the generalized Laguerre polynomial, is the electric-field amplitude, and is a normalization constant. The remaining quantities are the wavefront radius of curvature , the beam radius , the waist , the Gouy phase , and the Rayleigh range . In Eq. (22), we have also introduced a Gaussian pulse envelope satisfying . In particular, by using the integral , we find .
Near the focus, the field profile can be regarded as approximately invariant over . For and , we have , which exceeds all trapping lengths considered in the main text. We may therefore approximate . In this limit, we select the mode and , which near the propagation axis produces the potential energy
Equation (S3) defines a harmonic trapping potential with the weakly time-dependent frequency . An effective time-independent trapping frequency is obtained by integrating the potential in Eq. (23) over time and dividing by the effective interaction time , which gives . The field amplitude is related directly to the pulse energy through . In Table 1 in the main text, we report all the parameters required to produce a laser beam that traps the electrons with a trapping frequency of over a Rayleigh length of 20 mm and a pulse duration of . These are the parameters required to achieve the trapping described in the main text.
The cycle-averaged description derived above is valid under several conditions which are all met using a field amplitude and . Indeed, the quiver motion remains nonrelativistic as . Second, the quiver displacement is also small compared with the beam waist and any other characteristic length over which the envelope varies. Finally, aiming at producing trapping frequencies of the order of , we directly satisfy the secular condition .
Properties of a continuous electron beam confined in two dimensions
In this section, we describe the methods used to characterize an equilibrated electron beam confined in a two-dimensional trap. We first develop an analytical description within the harmonic approximation and then introduce the numerical approach used to determine the beam density and its spatial correlations.
Harmonic approximation in the one-dimensional electron chain phase
We consider a continuous electron beam confined by a transverse trap of frequency . Our aim is to characterize the Coulomb-ordered states formed by the electrons in the string and zig-zag structural phases. To this end, we consider electrons with linear density and write the Hamiltonian of the system as
We then expand the Hamiltonian to second order around the equilibrium positions , obtaining , where . The corresponding matrix is
where and the prime on the summation indicates that .
The equations of motion derived from the approximate Hamiltonian, , can be solved by expressing the displacement of the -th electron as a superposition of plane-wave modes, , where . Each mode is characterized by a wave vector , a frequency , and a polarization vector , with . The momentum can be expanded analogously as with . This mode expansion satisfies the equations of motion provided that
where we introduced the dynamical matrix and we used the property . Here, we impose periodic boundary conditions , which restrict the allowed values of to the first Brillouin zone. We also use the cyclic property and choose the origin at .
For a string configuration, the sum in Eq. (26) can be evaluated in the limit by placing the electrons at with . Using Eq. (25), we then obtain
with . Because is diagonal, the longitudinal and transverse mode frequencies follow directly from Eq. (26) and are given by and .
The structural transition from the string to the zig-zag configuration occurs when the transverse mode with wave vector becomes mechanically unstable. The instability condition is therefore , where we have used the relation and denotes the Riemann zeta function [9]. This condition is commonly expressed [10] in terms of the Wigner-Seitz radius and the normalized linear density as . Additional structural phases arise at higher densities, and their corresponding critical values of have been determined in previous studies [10].
We next examine the thermal stability of the string phase. For this purpose, it is convenient to express the displacements in terms of sine and cosine modes as , where and are real coefficients satisfying the relation . Similarly, the momenta can be written as , with . Using Eq. (26) together with the identity , the Hamiltonian reduces to the diagonal form .
By now taking the average of the mean-squared displacement along the chain, we compute the thermal mean-squared fluctuations as
where we computed the thermal average by using the transformations and that produce a Jacobian with unit determinant in the integration measure for in combination with the resulting Gaussian integrals . Here and in the rest of this section, thermal averages are defined as .
Using the longitudinal frequencies of the string phase, we obtain . The integral diverges at small arguments; for a finite number of particles, its asymptotic behavior is . It is well established that a strictly one-dimensional Coulomb crystal does not exhibit true long-range order when quantum fluctuations are included, even at zero temperature [11], or at finite temperature in the classical regime [12]. Nevertheless, substantial long-range ordering can emerge when the full three-dimensional motion is taken into account, as we demonstrate through the numerical simulations described below.
References
- [1]R. Zimmermann, M. Seidling, and P. Hommelhoff, Nat. Commun. 12, 390 (2021).
- [2]W. Paul, Rev. Mod. Phys. 62, 531 (1990).
- [3]P. K. Ghosh, Ion Traps (Clarendon Press, Oxford, 1995).
- [4]M. Kozák, T. Eckstein, N. Schönenberger, and P. Hommelhoff, Nat. Phys. 14, 121 (2018).
- [5]M. Kozák, N. Schönenberger, and P. Hommelhoff, Phys. Rev. Lett. 120, 103203 (2018).
- [6]M. Kozák, ACS Photonics 8, 431 (2021).
- [7]C. I. Velasco and F. J. García de Abajo, Phys. Rev. Lett. 134, 123804 (2025).
- [8]Y. Yang, Y. Li, and C. Wang, Journal of Optics 51, 910 (2022).
- [9]M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (Dover, New York, 1972).
- [10]R. W. Hasse and J. P. Schiffer, Ann. Phys. 203, 419 (1990).
- [11]H. J. Schulz, Phys. Rev. Lett. 71, 1864 (1993).
- [12]G. Morigi, J. Bollinger, M. Drewsen, D. Podolsky, and E. Shimshoni, Ion coulomb crystals: an exotic form of condensed matter (2025), arXiv:2508.07374 [physics.atom-ph].
