Relativistic electrons produced by foreshock disturbances

Foreshock disturbances -- large-scale (~1000 km to>30,000 km), transient (~5-10 per day - lasting ~10s of seconds to several minutes) structures [1,2] - generated by suprathermal (>100 eV to 100s of keV) ions [3,4] arise upstream of Earth's bow shock formed by the solar wind colliding with the Earth's magnetosphere. They have recently been found to accelerate ions to energies of several keV [5,6]. Although electrons in Saturn's high Mach number (M>40) bow shock can be accelerated to relativistic energies (nearly 1000 keV) [7], it has hitherto been thought impossible to accelerate electrons at the much weaker (M<20) Earth's bow shock beyond a few 10s of keV [8]. Here we report observations of electrons energized by foreshock disturbances to energies up to at least ~300 keV. Although such energetic electrons have been previously reported, their presence has been attributed to escaping magnetospheric particles [9,10] or solar events [11]. These relativistic electrons are not associated with any solar activity nor are they of magnetospheric origin. Further, current theories of ion acceleration in foreshock disturbances cannot account for electrons accelerated to the observed relativistic energies [12-17]. These electrons are clearly coming from the disturbances, leaving us with no explanation as to their origin.

Foreshock disturbances -large-scale (∼1000 km to >30,000 km), solitary (∼5-10 per day, transient (lasting ∼10s of seconds to several minutes)) structures 1,2 -generated by suprathermal (>100 eV to 100s of keV) ions 3,4 arise upstream of Earth's bow shock formed by the solar wind colliding with the Earth's magnetosphere. They have recently been found to accelerate ions to energies of several keV 5,6 . One type was found to have a distinct suprathermal electron population with energies >70 keV, which was attributed to a magnetospheric origin 7 . Although electrons in Saturn's high Mach number (M > 40) bow shock can be accelerated to relativistic energies (nearly 1000 keV) 8 , it has hitherto been thought impossible to accelerate electrons at the much weaker (M < 20) Earth's bow shock beyond a few 10s of keV 9 . Here we report observations of electrons energized by foreshock disturbances from 10s of eV up to at least ∼300 keV. We observe a single isotropic power-law from 100s of eV to 100s of keV, unlike previous studies 7 . All previous observations of energetic foreshock electrons have been attributed to escaping magnetospheric particles 7,10,11 or solar events 12 . We observe no solar activity and the single isotropic power-law cannot be explained by any magnetospheric source. Further, current theories of ion acceleration in foreshock disturbances cannot account for electrons accelerated to the observed relativistic energies [13][14][15][16][17][18] . These electrons are clearly coming from the disturbances, leaving us with no explanation for the acceleration mechanism.
We examine in detail particle velocity distribution functions measured by the low energy electron/total ion electrostatic analyzers 19 and high energy electron/total ion solid state telescopes 20 on the THEMIS spacecraft 21 near foreshock disturbances. Quasi-static vector magnetic fields (Bo) were measured by the fluxgate magnetometer 22 .
The disturbances occur in the ion foreshock 3,4 region upstream of the quasi-parallel bow shock, where the shock normal angle (θ Bn ) between the upstream quasi-static magnetic field (Bo) and the shock normal vector satisfies θ Bn < 45 • . We focused on observations near short large-amplitude magnetic structures 4 , hot flow anomalies 1 , and foreshock bubbles 2 . During the four orbital passes examined in detail, we identified 30 foreshock disturbances, 10 of which had clear energetic electron enhancements (five at short large-amplitude magnetic structures, two at hot flow anomalies, and three at foreshock bubbles). See Extended Data Fig. 1 for spacecraft orbits and Methods for further details of foreshock disturbances.
We observe energetic (≥30 keV) electron enhancements as short duration (∼10s of seconds to few minutes) enhancements in the electron fluxes above background by factors of ∼10-200 ( Fig. 1). They are localized to the large fluctuations in Bo (Figs 1a-1c) within the foreshock disturbances. The electron flux-time profiles for energies from ∼0.25 to >200 keV show no energy dispersion, i.e., fluxes increase at all energies simultaneously (Figs 1g-1i and 1m-1o). The low energy electron (Figs 1g-1i) and ion (Figs 1j-1l) data look qualitatively similar for disturbances with and without (Extended Data Fig. 2) energetic electron enhancements. Note that nearly all disturbances show enhanced energetic ion fluxes often to >300 keV (Figs 1p-1r). The ion enhancements will be examined in future work.
We then constructed pitch-angle -the angle between electron momentum and some reference vector, often Bo -distributions from the particle data to compare distributions during energetic electron enhancements (Figs 2a-2c) to those outside the enhancement periods (Figs 2d-2f ). The electron distributions are observed to be almost isotropic from ∼0.05-300 keV, with anisotropies (e.g., ratio of intensity or phase space density along Bo to that perpendicular to Bo) rarely exceeding factors of 2. Note that data above ∼140 keV are dominated by noise for the 2008-09-08 hot flow anomaly (Fig. 2b) and 2008-07-14 foreshock bubble (Fig. 2c) example enhancements. The data also show a power-law form with f (E) ∝ E −4 from as low as ∼0.25 keV up to highest energies observed during each enhancement (Figs 2a-2c). The distributions observed outside the enhancements (Figs 2d-2f ) show far more variability, only noise >12 keV, and in some cases significant anisotropies (Fig. 2f ).
Due to the high variability in Bo during the accumulation time of a single particle distribution we also constructed pitch-angle distributions by sorting the pitch angles with sub-spin period time resolution, using algorithms similar to previous work 23 , to remove artifacts of aliasing. We found that the isotropy was not a consequence of aliasing and is thus a real feature (see Extended Data Fig. 3).
To verify that the enhancements were not of solar origin, we examined the Wind and twin STEREO spacecraft radio data for solar radio bursts, which were not observed during the energetic electron enhancements observed by THEMIS. We also examined the Wind particle data for energy-dispersed profiles characteristic of solar energetic electrons 12 , which were also not observed. Finally, there were no interplanetary shocks, which can produce relativistic electrons 24,25 , observed by Wind during any of the electron enhancements. Further details excluding a solar source can be found in Methods and see Extended Data Fig. 4.
Next we ruled out the Earth's bow shock as a source by examining the following concepts. First, the two most viable shock acceleration mechanisms for electrons have the highest efficiencies in the quasi-perpendicular (θ Bn > 45 • ) region of the shock 9,16 . Second, due to a higher mobility along versus across Bo, any energetic electron distribution observed several Earth radii (i.e., tens to hundreds of Larmor radii) away from the source would be highly anisotropic along Bo, as previously reported 26 , inconsistent with our observations (e.g., Figs 2a-2c). Third, the maximum observed energies are at most several tens of keV energies 24,25,27 seen as magnetic field-aligned beams at the electron foreshock edge, whereas we observe isotropic distributions up to at least ∼300 keV within the ion foreshock (Figs 1 and 2). Finally, both of these mechanisms should produce some energy dispersion for upstream observations, which we do not observe (Fig. 1). For a detailed discussion of these mechanisms see Methods and Extended Data Fig. 5.
Another possibility is that the enhancements arise as a "pile up" of particles at the magnetic gradients, which act like mirrors. However, we do not observe energydispersed profiles within the magnetic compressions (i.e., due to energy-dependent Larmor radii effects and diffusion) or enhanced intensities outside the foreshock disturbances or pitch-angle distributions peaking perpendicular to Bo. Further, the Larmor radius of a ∼90 keV electron, in the range of magnetic field magnitudes observed, span ∼26-1080 km. These values are comparable to and larger than the gradient scale length of most collisionless shock waves 4 and comparable to the scale size of some foreshock disturbances, which precludes adiabatic reflection between the two merging shocks 18 . Thus, we can rule out the "pile up" scenario as a source.
Previous reports of energetic electrons in the ion foreshock attributed the enhancements to a magnetospheric source 10,11,13 . One study 7 found energetic electrons (∼70-200 keV) within hot flow anomalies that were nearly isotropic, but the authors explicitly stated that the energetic electrons were a separate population from those at lower energies and that they were of magnetospheric origin. There are multiple differences between these results and our observations, including: (1) while hot flow anomalies can be magnetically connected to the magnetosheath, the other two foreshock disturbances in our study are not; (2) we observe a single power-law from 100s of eV to ≥140 keV in many enhancements, indicative of a common acceleration mechanism; and (3) the combination of a single power-law and an isotropic distribution over such a large range of energies cannot be explain by a magnetospheric source.
Finally, the enhancements were often observed with geomagnetic activity 10,11,13,28 and exhibited large anisotropies (ratios of sunward-to-anti-sunward fluxes 29 up to 5:1 and field-aligned-to-perpendicular fluxes 11 up to 6:1). Our observations are not consistent with any of these studies for the following reasons. First, since our electron distributions are nearly isotropic, they do not exhibit either highly field-aligned ( Fig. 2) nor antiearthward intensity anisotropies (see Extended Data Fig. 5). Second, we do not observe enhanced geomagnetic activity before or during any of the energetic electron enhancements (see Extended Data Fig. 6). Finally, the enhancements are always isolated within the foreshock disturbances, thus not consistent with escaping magnetospheric particles. For further details, see Methods.
No previous work ever considered that foreshock disturbances could locally produce electrons to these energies. It is not clear why only some disturbances produce enhancements nor can current theory explain the electron energization, but this is beyond the scope of this study.
The electron energization is not occurring locally at the main shock, but remotely. The most outstanding unanswered question in shock acceleration theory is the so called "injection problem" (i.e., how to get thermal particles up to suprathermal energies before they are convected downstream), where previous work has only considered local energization at the shock. Therefore, these observations provide a new avenue through which electrons can be non-locally pre-energized to high enough energies to undergo further acceleration when interacting with astrophysical shocks. Given the ubiquity of foreshocks upstream of collisionless astrophysical shocks, we expect foreshock disturbances to be ubiquitous as well, which could fundamentally change our understanding of collisionless shocks.
Online Content Methods, along with any additional Extended Data display items and Source Data, are available in the online version of the paper; references unique to these sections appear only in the online paper.

Methods
Foreshock disturbance properties. Since previous simulation 30 and observational 23,31 studies found that magnetosonic-whistler waves 4 -right-hand polarized electromagnetic plasma waves, with density fluctuating in phase with |Bo| -can accelerate particles to >1 keV, we limited the types of foreshock disturbances to those having a magnetosonic-whistler nature. We further limited our search to disturbances occurring multiple times per day, those with nonlinear properties, and those capable of forming shock waves, all properties associated with energizing particles. Therefore, the three foreshock disturbances we examined, short large-amplitude magnetic structures 4,6,32-35 , hot flow anomalies 1,36-41 , and foreshock bubbles 2,42,43 , are all produced by the interaction between the incident solar wind and suprathermal (>100 eV to 100s of keV) ions 3,4,44 . Short largeamplitude magnetic structures are short duration (∼few to 10s of seconds), nonlinear large amplitude (δB/B > 2), monolithic "magnetic pulsations" with spatial scales of ∼1000 km that can exhibit a soliton-like behavior (i.e., large amplitude fluctuations are fast and spatially narrow) 33,35 . Both hot flow anomalies and foreshock bubbles are localized rarefaction regions surrounded by compression regions that are effectively "carved out" by an accumulation of suprathermal ions along a discontinuity in the interplanetary magnetic field. The difference is that the compression regions for hot flow anomalies are centered on the discontinuity and the discontinuity must interact with the Earth's bow shock, whereas foreshock bubbles form upstream of the discontinuity and the discontinuity need not interact with the Earth's bow shock 1,2 . Both hot flow anomalies and foreshock bubbles are several Earth radii in scale (i.e., >10,000 km). Exclusion of solar source. We first eliminated interplanetary shocks as a possibility by examining the Wind shock database at the Harvard Smithsonian for Astrophysics (Online at http://themis.ssl.berkeley.edu/index. shtml) finding no interplanetary shocks during any of the energetic electron enhancements.
We next examine the radio data from the Wind and STEREO spacecraft (Extended Data Fig. 4). There are no clear radio bursts or any evidence of significant radio activity on the sun during any of the four THEMIS foreshock passes. For comparative purposes, we include a date with clear solar radio bursts (Extended Data Figs 4e, 4j, and 4o). The enhanced radio intensity near ∼200 kHz (Figs 4f -4i) is most likely auroral kilometric radiation 45 , which would have no effect on particle observations by THEMIS. Examination of the solid state telescope particle data from Wind 46 (freely available on CDAWeb, see Data availability below) show no significant energetic electron enhancements during any of the enhancements observed by THEMIS. Finally, the electron data show no evidence of forward energy dispersion (i.e., higher energies arrive before lower due to a time-of-flight effect) characteristic solar energetic electrons 12 ( Fig. 1 and Extended Data Figs 5k-5n). Note that the energetic ions exhibit slightly larger anisotropies than the electrons with antiearthward intensities generally dominating (see Extended Data Fig. 5). Exclusion of Earth's bow shock as source. The most common shock acceleration mechanisms cited are diffusive shock acceleration 15 , shock drift acceleration 13,16 , and the "fast Fermi" mechanism 9,47 . However, for electron acceleration at the Earth's bow shock we can rule out these mechanisms for the following reasons. First, though diffusive shock acceleration predicts an isotropic particle distribution, it is more efficient for quasi-parallel (θ Bn < 45 • ) shocks with pre-existing upstream electromagnetic fluctuations and the efficiency increases with particle kinetic energy 14,17 . This mechanism also cannot energize electrons below ∼100 keV because their Larmor radii are smaller than the gradient scale lengths of the shock ramp 48,49 and upstream electromagnetic fluctuations 4 . Further, this mechanism predicts an inverse energy dispersion (i.e., lower energies enhance first) 11,29 , which is not observed in the energetic electron enhancements ( Fig. 1 and Extended Data Fig. 5). Thus, diffusive shock acceleration is generally ignored as a mechanism for energizing electrons from thermal to relativistic energies at the Earth's bow shock.
Second, shock drift acceleration predicts anisotropic velocity distributions, perpendicular downstream of shock and field-aligned far upstream of the Earth's bow shock 50,51 , neither of which are observed ( Fig. 2 and Extended Data Fig. 5). The mechanism efficiency decreases with increasing ratio of shock speed to cos θ Bn relative to the particle thermal energy because this increases the minimum energy threshold requirement 52 , where each interaction with the shock can produce energy gains of factors >10 for strong quasi-perpendicular shocks 53 . However, any electron distribution observed far upstream (e.g., near the foreshock disturbances) would be highly anisotropic along the magnetic field streaming away from the shock as previously observed 26,54 , inconsistent with the isotropic distributions we observe ( Fig. 2 and Extended Data Fig. 5). Thus, we can rule out shock drift acceleration at the Earth's bow shock as a source.
Third, fast Fermi acceleration assumes electrons undergo a single adiabatic reflection -particle conserves its magnetic moment, µ = mev ⊥ 2 /2Bo ∼ constant, during the reflection, where me is the electron mass and v ⊥ is the speed perpendicular to Bo -and gains energy proportional to the shock speed divided by cos θ Bn 9,55 . Previous studies 26,47,54 proposed this mechanism as an explanation for the "thin sheets" of highly anisotropic (i.e., field-aligned streaming away from Earth's bow shock) energetic electrons. To satisfy the condition µ ∼ constant, the magnetic gradient scale length must be larger than the particle Larmor radius. Further, for significant electron energy gains this mechanism requires either very large shock speeds compared to typical electron thermal speeds (i.e., ∼1500-3000 km/s) or θ Bn 88 • . Since the Earth's bow shock is very slow (i.e., typically ∼100-500 km/s), the Larmor radii of electrons few hundred eV are comparable to the shock ramp thickness 48,49 , and this mechanism can only energize electrons with large pitch-angles, fast Fermi acceleration is not expected to produce energies beyond several 10s of keV 9,25,27,55 . We thus rule out the Earth's bow shock as the source of these energetic electron enhancements. Exclusion of Earth's magnetosphere as source. As discussed in the main article, the observed energetic electron distributions are not highly anisotropic along the magnetic field streaming away from the Earth as previously reported in studies arguing for a magnetospheric source 10,11,28,50,56 . These studies observed enhanced geomagnetic activity in association with bursts of energetic electrons, where they argued that substorms -a fundamental mode of the terrestrial magnetosphere resulting in magnetospheric circulation/flows and enhanced auroral activity 21 -led to an increased rate of magnetospheric "leakage." During substorms geostationary spacecraft can observe intense and rapid changes in 10s of keV to MeV electron fluxes 21 . One measure of substorm activity can be given by the well known AE indices 57 . These previous studies defined enhanced geomagnetic activity as an AE index > 200 nT.
Extended Data Fig. 6 shows the AE indices with colorcoded bars indicating the time ranges for the example foreshock disturbances in Fig. 1. The AE index (Extended Data Figs 6q-6t) was < 200 nT for >1 hour prior to and during the three example disturbances with energetic electron enhancements (Figs 1m-1o). Thus, we can rule out a magnetospheric source due to the inconsistent pitch-angle distributions and lack of enhanced geomagnetic activity.
One previous study 7 did find nearly isotropic energetic electrons (∼70-200 keV) within hot flow anomalies. However, the authors explicitly state these are not a suprathermal tail of the thermal electrons and that they originated from the magnetosphere. There are several important differences with our results: (1) we observe a single power-law from 100s of eV to ≥140 keV in many enhancements, suggesting a common acceleration mechanism; (2) we always observe ions above ∼10 keV with every foreshock disturbance; (3) we observe most electron enhancements (8/10) within short large-amplitude magnetic structures and foreshock bubbles, both of which are disconnected from the bow shock; and (4) the "mag-netic bottle" model proposed to explain isotropy in the previous study 7 would not work for short large-amplitude magnetic structures because they contract as they evolve, not expand. Instrument details. Quasi-static (i.e., finite gain from zero up to Nyquist frequency) vector magnetic field measurements (Bo) were obtained using the fluxgate magnetometer 22 at 4 and 128 samples per second. The data are presented in units of nanotesla [nT] in the geocentric solar ecliptic coordinate basis.
Particle data are stored as velocity distribution functions covering 4 π steradian over an energy range defined by instrument design. The electrostatic analyzers 19,58 , onboard each THEMIS 21 spacecraft, detect particles using anodes placed behind microchannel plate detectors and cover an energy range of few eV to over 25 keV. The number and placement of the anodes determines the poloidal/latitudinal angular resolution, which is usually ∆θ ∼ 22.5 • for both the electron and ion electrostatic analyzers (Note that ∆θ can be as low as ∼5 • in some ion instrument modes.). The azimuthal/longitudinal resolution, ∆φ, is limited by the spacecraft spin rate and instrument design and mode of operation, but is generally ∼11.25 • . The energy resolution, ∆E/E, is defined by the instrument design and mode of operation but is generally ∼20% for both electrostatic analyzers.
At higher energies (i.e., ∼30 keV to over 500 keV), data from the electron and ion solid state telescopes 20,59 onboard each THEMIS spacecraft were used. Each detector is comprised of two identical telescopes mounted at different angles on the side of the spacecraft body 21 . The angular and energy resolution is usually ∆θ ∼ 30 • -40 • , ∆φ ∼ 22.5 • , and ∆E/E ∼ ∼ 30%. For instance, the ∼293 keV energy bin actually includes energetic electrons with ∼249-337 keV energies.
All particle data presented herein, except for the high energy ions, were taken while in burst mode, which has a time resolution equal to the spacecraft spin period (i.e., ∼3 seconds). Even though the high energy ion data were measured in a different mode, the time resolution is still the spin period for the intervals presented. Unit conversion. The raw data are measured in units of counts, which correspond to the number of events with a pulse height exceeding a defined threshold specific to each instrument. Conversion to intensity and/or phase space density requires knowledge of the instruments efficiency 60,61 , deadtime 62,63 , accumulation time, and optical geometric factor [64][65][66] .
Particle intensity is defined as: number of particles, per unit area, per unit solid angle, per unit time, per unit energy (e.g., # cm −2 s −1 sr −1 eV −1 ). This unit is not a Lorentz invariant, thus it requires one taking into account the Compton-Getting effect. Phase space density is defined as: number of particles, per unit spatial volume, per unit "velocity volume" (e.g., # s +3 cm −3 km −3 ). This unit is a Lorentz invariant under conditions when phase space is incompressible (i.e., when Liouville's theorem reduces to df /dt = 0), which is true for most cases in in situ space observations 65 .
The exact details of the unit conversion can be found in the THEMIS calibration software, called SPEDAS, found at: http://themis.ssl.berkeley.edu/index.shtml. Reference frames and coordinate systems. All particle data shown herein has been transformed from the spacecraft to the bulk flow reference frame using a relativistically correct Lorentz transformation. The distributions were converted into units of phase space density prior to any frame transformation, thus any anisotropies due to the Compton-Getting effect 67,68 have been removed. Reference frame transformations were performed through the following steps: 1. bulk flow velocities were determined from the first velocity moment of the low energy (<30 keV) ion velocity distributions 69 ; 2. all particle distributions were converted from raw counts to phase space density; 3. particle distributions were then transformed into the bulk flow rest frame using a standard Lorentz velocity transformation; 4. the particle distributions shown in units of intensity were converted to phase space density prior to the transformation and then back to intensity. Direction-dependent spectra, as opposed to omnidirectional averages (i.e., average over all solid angles), called pitch-angle distributions were calculated through the following steps: 1. particle distributions were transformed into the bulk flow rest frame as described above; 2. construct particle velocity unit vector for the m th particle distribution,v i,j,k m , for the k th energy bin from the i th latitude and j th longitude detector look directions described in Instrument details above; 3. define the projection angle, α i,j,k m , betweenv i,j,k m and the respective orientation unit vector,û m (e.g., Bo/|Bo|), at the measurement time of the m th particle distribution, where α i,j,k m = cos −1 (v i,j,k m ·û m ); and 4. define three cuts for the m th particle distribution by averaging data within ±22.5 • of the parallel, perpendicular, and anti-parallel directions defined byû m . The two relevant directions about which we oriented the particle distributions are the local quasi-static magnetic field vector, Bo, and the spacecraft-to-Earth unit vector, e SC , at the time of each distribution. Note that the formal definition of a pitch-angle distribution is constructed only with respect to Bo but we use the term here for both orientations for brevity.
The high time resolution equivalent of the above algo-rithm involves only a few differences described below. 1. Instead of using a singleû m for the m th particle distribution, we defineû i,j,k m for the k th energy bin from the i th latitude and j th longitude detector look directions of the m th particle distribution. 2. Now the pitch-angles are defined as α i,j,k m = cos −1 (v i,j,k m ·û i,j,k m ). The result is a pitch-angle distribution with fewer aliasing effects due to the use of a singleû m averaged over the duration of the m th particle distribution (Extended Data  Fig. 3).
The exact details of the rotations and frame transformations can be found in the additional analysis software at: https://github.com/lynnbwilsoniii/wind 3dp pros. Particle data presentation. All particle distributions are presented in the bulk flow rest frame in a physically significant coordinate basis, e.g., magnetic field-aligned coordinates. We define the bulk flow reference frame as described above (in Reference frames and coordinate systems). The energy ranges listed above (in Instrument details) are the measured midpoint kinetic energies in the spacecraft frame of reference. Energy values will change under any Lorentz transformation. Thus, the energies for the ion electrostatic analyzer data are not constant in time and show a large variability owing to the large variability of the bulk flow near foreshock disturbances. In contrast, the low energy electron data above ∼50 eV and all solid state telescope data suffer little energy change under these Lorentz transformations, thus can be approximated as constant in time. Therefore, we show the low and high energy electron and high energy ion data as stacked line plots vs. time where each line corresponds to a different energy and the low energy ion data are presented as a dynamic energy spectrogram of energy vs. time with the color scale indicating the particle intensity. Note that while the electron solid state telescope can measure electrons to >400 keV, for only the 2008-08-19 event in Fig. 1 were significant fluxes observed in energy bins >140 keV. Data availability. The THEMIS data used in this paper is publicly available at: http://themis.ssl.berkeley.edu/index.shtml. The Wind and STEREO radio data were taken from the S/WAVES website at: http://swaves.gsfc.nasa.gov/data access.html. Solar wind data was taken from the OMNI data products found on CDAWeb at: http://cdaweb.gsfc.nasa.gov. The Wind interplanetary shock list can be found at: https://www.cfa.harvard.edu/shocks/wi data/. Code availability. The THEMIS instrument calibration software, called SPEDAS, can be found at: http://themis.ssl.berkeley.edu/index.shtml; and additional analysis software can be found at: Example foreshock disturbances with electron enhancements   Figure 2| Example pitch-angle spectra inside and outside electron enhancement periods for each of the example disturbances in Fig. 1. One-dimensional directional cuts (with respect to Bo) of the merged particle velocity distributions from the low and high energy electron data, in units of phase space density [# s +3 cm −3 km −3 ], on log-log plots with uniform horizontal and vertical axis ranges of ∼10 −26 -10 −12 s +3 cm −3 km −3 and ∼0.050-400 keV, respectively. The color-coded boxes above each plot correspond to the color-coded arrows in Fig. 1. Each panel shows cuts parallel (red), perpendicular (green), and anti-parallel (blue) to Bo, where missing data points indicate an absence of significant flux. The orange dashed line shows a power-law (defined in a) for perspective. a-c, distributions during the peak energetic electron enhancements in Fig. 1. d-f, distributions during periods of inactivity in Fig. 1. See Methods for details on the calculation of these quantities.   Example foreshock disturbances without electron enhancements XYZ Extended Data Figure 2| Three example foreshock disturbances without energetic electron enhancements. This figure has the same format as Fig. 1 with magnetic fields shown in nT and particle data in intensity [# cm −2 s −1 sr −1 eV −1 ]. a-c, |Bo| [nT] at 4 samples per second. d-f, Bo [nT] at 4 samples per second. g-i, low energy electron (i.e., ∼50 eV to ∼12 keV) intensity. j-l, low energy ion (i.e., ∼10 eV to ∼25 keV) intensity. m-o, high energy electron (i.e., ∼30-300 keV) intensity. p-r, high energy ion (i.e., ∼30-430 keV) intensity. See Methods for details on the calculation of these quantities.  Figure 3| Comparison of a single particle distribution constructed from low and high time resolution fields. The pitch-angle distribution corresponds to that in Fig. 2c with the same format. a, pitch-angles constructed using ∼128 samples per second Bo. b, pitch-angles constructed using ∼3 second averaged Bo. See Methods for details on the calculation of these quantities.   Figure 4| Wind and STEREO spacecraft daily radio spectra plots. Radio wave spectra, from the Wind 71 (panels f -j) and STEREO 72 Ahead (panels a-e) and Behind (panels k-o) spacecraft, shown as frequency vs. time plots of daily timeand frequency-averaged spectral intensity plots of decibels above background. The frequencies shown range from ∼10 kHz to ∼10 MHz. e, j, o, show an example radio burst occurring shortly after 12:00 UT on 2011-10-21 seen by all three spacecraft. Comparison with the example radio burst clearly shows that there is insigificant solar activity for the four foreshock crossings examined in this study.  . c-f, low energy electron (i.e., ∼50 eV to ∼12 keV) intensity. g-j, low energy ion (i.e., ∼10 eV to ∼25 keV) intensity. k-n, high energy electron (i.e., ∼30-140 keV) intensity. o-r, high energy ion (i.e., ∼30-430 keV) intensity.