Atmosphere-Breathing Electric Propulsion (ABEP)

• Atmosphere-Breathing Electric Propulsion (ABEP) is an electric propulsion method that collects residual atmospheric particles to use as propellant for drag compensation in very low orbits.
• It integrates specialized intakes with optimized gas–surface interaction models and RF thruster architectures, achieving capture efficiencies up to 94.3% under varying orbital conditions.
• Recent developments incorporate data-driven scattering kernels and active cryogenic intake concepts to enhance thrust efficiency and address uncertainties in rarefied flow and material erosion.

Atmosphere-Breathing Electric Propulsion (ABEP) is an electric-propulsion architecture in which a spacecraft intake collects residual atmospheric particles encountered along the ram direction and delivers them to a thruster, so that the ambient atmosphere itself serves as propellant for continuous drag compensation in very low orbital regimes. In the literature surveyed here, ABEP is primarily formulated for Very Low Earth Orbit (VLEO, below about $450$ km) and, by extension, for very low orbits around other atmospheric bodies. The system is usually decomposed into an intake, a compression or flow-conditioning stage, an ionization stage, and an acceleration stage, with feasibility governed by the coupled balance among rarefied-flow capture, thruster efficiency, and spacecraft drag (Romano et al., 2021, Vaidya et al., 2022, Romano, 2 Jul 2026).

1. Concept and orbital regime

ABEP is motivated by the fact that at very low altitudes the residual atmosphere generates substantial drag, while conventional electric propulsion remains constrained by finite onboard propellant. In the VLEO regime, the flow is free molecular or highly rarefied, with $Kn > 10$ emphasized in recent gas–surface interaction work, so the dominant physics is not continuum compression but wall-mediated transport, scattering, and backflow. Atmospheric particles strike spacecraft surfaces at hypersonic orbital speeds of $6{,}000$–$10{,}000$ m/s; representative values cited across the ABEP literature are $u_\infty \approx 7.8$ km/s, $v \approx 7.5$–$7.8$ km/s, and $T_\infty$ in the several-hundred-Kelvin range (Schütte et al., 30 Jun 2026, Parodi et al., 17 Apr 2025).

The ambient composition is altitude dependent. For Earth VLEO, the cited models and simulations include atomic oxygen $O$, $N_2$, $Kn > 10$0, He, Ar, H, and N, with $Kn > 10$1 and $Kn > 10$2 usually dominating mass flow; recent intake and scattering studies also treat $Kn > 10$3 formation through surface recombination and explicitly note the erosive role of atomic oxygen in materials selection (Romano et al., 2021, Parodi et al., 17 Apr 2025). The same ABEP logic is extended in mission studies to Mars, where very low Mars orbit is analyzed with $Kn > 10$4-dominated atmospheres (Romano et al., 2021, Vaidya et al., 2022).

At system level, the defining balance is between aerodynamic drag and propulsive thrust. The recurring relations are

$Kn > 10$5

and

$Kn > 10$6

with stationkeeping requiring $Kn > 10$7 (Romano et al., 2021, Vaidya et al., 2022, Romano, 2 Jul 2026). Because $Kn > 10$8 scales with density and frontal area, and $Kn > 10$9 scales with intake capture, ABEP feasibility is intrinsically coupled to intake geometry, gas–surface interaction, and available power rather than to thruster performance alone. Earth-focused mission analyses place a practical lower bound near $6{,}000$0 km because of heating and re-entry onset, while upper ABEP design envelopes such as $6{,}000$1–$6{,}000$2 km are chosen because at higher altitude collectible mass flow becomes marginal for typical spacecraft (Romano et al., 2021, Romano et al., 2022).

2. Intake physics, capture, and gas–surface interaction

The intake is the defining ABEP subsystem because it must collect a highly collimated, hyperthermal beam and deliver it to the thruster under free-molecular conditions. The standard intake performance measure is the collection efficiency

$6{,}000$3

with the delivered mass flow

$6{,}000$4

A closely related free-stream estimate is $6{,}000$5 (Romano et al., 2021, Romano et al., 2021, Romano et al., 2022).

The reviewed intake literature repeatedly treats gas–surface interaction through idealized limiting models. In the PICLas intake-design study, Maxwell boundary conditions are implemented with MomentumACC = 0.0 for specular reflection and MomentumACC = 1.0 for diffuse reflection (Romano et al., 2021). In a separate intake-thruster design paper, energy exchange is described through

$6{,}000$6

with the paper stating that specular reflection corresponds to $6{,}000$7 and diffuse reflection to $6{,}000$8 (Romano et al., 2022). The reviewed papers therefore do not use a single accommodation-coefficient convention, but they agree on the physical distinction: specular walls preserve directionality and energy, whereas diffuse walls thermalize and randomize the reflected flux.

That distinction leads directly to the observed geometry classes. Diffuse intakes in the DISCOVERER studies use hexagonal honeycomb front ducts followed by a conical chamber; the ducts act as molecular traps, larger central ducts increase direct transmission, and narrower outer ducts suppress perimeter escape. Under these assumptions, the D-MDAR concept at $6{,}000$9 km yields $10{,}000$0 m$10{,}000$1, $10{,}000$2 m$10{,}000$3, $10{,}000$4 s$10{,}000$5, $10{,}000$6 s$10{,}000$7, $10{,}000$8, $10{,}000$9 mg/s, and $u_\infty \approx 7.8$0 Pa, with $u_\infty \approx 7.8$1 decreasing to $u_\infty \approx 7.8$2 by $u_\infty \approx 7.8$3 km (Romano et al., 2021). Specular intakes instead use parabolic scoops that optically focus incident trajectories into the discharge channel; the S-FT design at $u_\infty \approx 7.8$4 km reaches $u_\infty \approx 7.8$5 m$u_\infty \approx 7.8$6, $u_\infty \approx 7.8$7 s$u_\infty \approx 7.8$8, $u_\infty \approx 7.8$9 s$v \approx 7.5$0, $v \approx 7.5$1, $v \approx 7.5$2 mg/s, and $v \approx 7.5$3 Pa (Romano et al., 2021).

Misalignment sensitivity is also largely a gas–surface problem. For the diffuse D-MDAR intake at $v \approx 7.5$4 km, $v \approx 7.5$5 falls from $v \approx 7.5$6 at $v \approx 7.5$7 to $v \approx 7.5$8, $v \approx 7.5$9, $7.8$0, and $7.8$1 at $7.8$2, $7.8$3, $7.8$4, and $7.8$5, respectively (Romano et al., 2021). The specular S-FT intake maintains nearly constant $7.8$6 up to about $7.8$7, then drops sharply; by about $7.8$8, $7.8$9 (Romano et al., 2021). The literature explicitly notes that real surfaces deviate from ideal specular and diffuse limits because roughness, contamination, and atomic-oxygen erosion drive reflection behavior over time (Romano et al., 2021).

A further refinement appears in recent molecular-dynamics-based work on gas–surface scattering for ABEP intakes. There the central object is the scattering kernel $T_\infty$0, defined in the local frame by

$T_\infty$1

with normalization, positivity, and detailed-balance constraints imposed so that multi-bounce deceleration can be modeled from thermal to hypersonic velocities (Schütte et al., 30 Jun 2026). This is particularly important because ABEP intakes rely on successive wall collisions to decelerate particles from orbital speed to near-thermal velocity before they reach the plasma stage.

3. Thruster architectures and intake–thruster coupling

The thruster side of ABEP is dominated by electrodeless RF concepts because chemically aggressive species, especially atomic oxygen, erode electrodes, grids, and discharge channels in conventional electric propulsion. The RF Helicon-based Inductive Plasma Thruster (IPT) developed at IRS within the DISCOVERER project is explicitly designed as a contactless, neutralizer-free device: an intake feeds neutrals into a discharge channel, an RF antenna and axial magnetic field sustain a helicon discharge, and the exhaust is a quasi-neutral plasma plume (Romano et al., 2020, Romano et al., 2022, Romano, 2 Jul 2026).

The laboratory IPT reported by IRS uses a birdcage antenna resonant at $T_\infty$2 MHz, an external solenoid capable of $T_\infty$3 mT, a vacuum chamber with base pressure $T_\infty$4 Pa, and operation on Ar, $T_\infty$5, and $T_\infty$6. The design is centered on resonance and low reflected power; in the 2026 dissertation, the finalized birdcage has measured $T_\infty$7 dB at $T_\infty$8 MHz with $T_\infty$9, implying $O$0 coupling in vacuum, and post-test measurements improved to $O$1 dB with $O$2 (Romano, 2 Jul 2026). Experimentally, successful ignition and operation across relevant mass flows were demonstrated with low power consumption, reported as $O$3 W or $O$4 W depending on the study (Romano et al., 2022, Romano, 2 Jul 2026).

A second IRS line of work uses the inductively heated IPG6-S as an ABEP-relevant test-bed for an Inductive Plasma Thruster. In that configuration the RF source operates at about $O$5 MHz with active RF power up to about $O$6 kW, using $O$7 and air as working gases. The measured electric-to-thermal coupling efficiency is reported up to about $O$8 for $O$9 and about $N_2$0 for air, with a maximum specific enthalpy $N_2$1 MJ/kg at $N_2$2 mg/s. Under the paper’s upper-limit assumption that the measured enthalpy is fully converted to directed kinetic energy,

$N_2$3

which gives $N_2$4 km/s and an upper-limit thrust of about $N_2$5–$N_2$6 mN at low altitudes (Romano et al., 2021).

A distinct particle-based simulation of a CubeSat-scale ABEP system resolves the intake and thruster jointly with DSMC for neutrals and PIC-MCC for the plasma. In that model, the RF coil is driven at $N_2$7 MHz, the screen grid is RF biased with $N_2$8 V and $N_2$9 V, the absorbed RF power stabilizes near $Kn > 10$00 W, and the steady coil current is about $Kn > 10$01 A. The simulated ion exhaust speed for $Kn > 10$02 under $Kn > 10$03–$Kn > 10$04 V is $Kn > 10$05–$Kn > 10$06 km/s, corresponding to $Kn > 10$07–$Kn > 10$08 s (Parodi et al., 17 Apr 2025). That study separates the architecture into intake, ionization stage, acceleration grids, and neutral exhaust, while explicitly noting that one-way coupling neglects neutral depletion and back-pressure feedbacks.

A related but distinct electrodeless RF concept is the SUb-atmospheric Radio-frequency Engine (SURE), designed for near-space pressures of $Kn > 10$09–$Kn > 10$10 Pa rather than VLEO free-molecular conditions. SURE uses two axially separated solenoid coils at $Kn > 10$11 MHz and up to $Kn > 10$12 kW, combining inductive coupling under each leg with capacitive coupling between the legs. It reports plasma density up to $Kn > 10$13 m$Kn > 10$14 and electron temperature up to $Kn > 10$15 eV without plasma-facing electrodes or external magnetic field (Yuan et al., 2020). Although its target regime differs from orbital ABEP, it belongs to the broader class of air-breathing electrodeless RF propulsion schemes.

4. Modeling hierarchy: DSMC, PIC, and learned scattering kernels

ABEP has been developed through a layered modeling hierarchy because no single solver spans free-molecular intake transport, gas–surface scattering, plasma production, and mission-level drag compensation. On the neutral side, DSMC is the dominant intake tool. PICLas is used in the DISCOVERER intake-design work for free-molecular or highly rarefied intake transport with Maxwell wall models (Romano et al., 2021). Pantera is used for a one-way coupled DSMC$Kn > 10$16PIC-MCC workflow in a full 3D CubeSat-sized ABEP geometry, where DSMC delivers the neutral density to the plasma solver on the same unstructured mesh (Parodi et al., 17 Apr 2025). A hybrid PIC–DSMC framework has also been used to analyze ground testing of ABEP intakes under electric-propulsion plumes, including ion-surface neutralization and sputtering (Moon et al., 22 Sep 2025).

Recent work extends this hierarchy by replacing idealized wall models with a data-driven scattering kernel learned from molecular dynamics. The 2026 scattering-kernel paper parameterizes the conditional density $Kn > 10$17 with a conditional Real-valued Non-Volume Preserving flow, cRealNVP, and enforces positivity and normalization by construction. The kernel is trained on molecular-dynamics data for atomic oxygen on amorphous $Kn > 10$18, with incident velocities ranging from thermal speeds to $Kn > 10$19 m/s and nine polar angles $Kn > 10$20 (Schütte et al., 30 Jun 2026).

The thermodynamic constraint central to that work is detailed balance. In its probability form,

$Kn > 10$21

and in the flux-weighted form,

$Kn > 10$22

with

$Kn > 10$23

The total loss is

$Kn > 10$24

where $Kn > 10$25 penalizes deviations from detailed balance on equilibrium samples (Schütte et al., 30 Jun 2026). This matters operationally because ABEP intake particles undergo multiple collisions and eventually enter near-thermal regimes where equilibrium consistency is no longer optional.

Validation results are explicit. When a Maxwell flux at $Kn > 10$26 K is scattered with cRealNVP, reservoir simulation starting from $Kn > 10$27 K converges to a mean of about $Kn > 10$28 K over the last $Kn > 10$29 steps, corresponding to about $Kn > 10$30 error. The prior CVAE-based kernel stabilizes at about $Kn > 10$31 K, which the paper presents as evidence of equilibrium inconsistency (Schütte et al., 30 Jun 2026). The cRealNVP model also matches molecular-dynamics reflected distributions across incident magnitudes $Kn > 10$32, $Kn > 10$33, and $Kn > 10$34 m/s and extrapolates successfully to $Kn > 10$35 m/s, while flat-plate DSMC tests show that both cRealNVP and CVAE reproduce MD-derived lift, drag, and $Kn > 10$36 trends better than a fixed-accommodation Maxwell model (Schütte et al., 30 Jun 2026).

Because the learned $Kn > 10$37 is explicit over the full speed range, it can be embedded directly in multi-bounce intake simulation. The reported workflow initializes particle state at the intake entrance, propagates ballistically to the next wall, samples outgoing velocity from $Kn > 10$38, repeats over successive wall interactions, and terminates when the particle is captured through the throat, lost to free stream, sticks if that is modeled, or falls below a thermal threshold (Schütte et al., 30 Jun 2026). This formulation connects molecular-scale scattering directly to capture efficiency, compression, and the phase-space distribution entering the ionization stage.

5. Performance envelopes, active intakes, and mission applications

ABEP performance is strongly assumption dependent because altitude thresholds move with $Kn > 10$39, $Kn > 10$40, $Kn > 10$41, $Kn > 10$42, and available electrical power. In the passive-intake DISCOVERER study, the mass flow and required exhaust velocity already show a large specular–diffuse separation. At $Kn > 10$43 km, the S-FT specular intake provides $Kn > 10$44 kg/s, while D-MDAR gives $Kn > 10$45 kg/s. For full drag compensation with the assumed frontal areas, the required exhaust velocity is about $Kn > 10$46–$Kn > 10$47 km/s for S-FT over $Kn > 10$48–$Kn > 10$49 km, but about $Kn > 10$50–$Kn > 10$51 km/s for D-MDAR (Romano et al., 2021). This is why the same paper concludes that the specular parabolic scoop with focus inside the discharge channel is the most promising intake in its design set.

Mission-level Earth analyses confirm that conclusion under broader spacecraft assumptions. In the Earth circular-orbit study with $Kn > 10$52 m$Kn > 10$53, $Kn > 10$54, and $Kn > 10$55, a $Kn > 10$56 km orbit requires $Kn > 10$57 km/s and peak power $Kn > 10$58 kW for $Kn > 10$59, whereas $Kn > 10$60 reduces these to $Kn > 10$61 km/s and about $Kn > 10$62 kW (Vaidya et al., 2022). The same study finds that full drag compensation at $Kn > 10$63 km is feasible with $Kn > 10$64 kW for $Kn > 10$65 and $Kn > 10$66 in the range $Kn > 10$67–$Kn > 10$68, while at $Kn > 10$69 km feasibility at that power level requires reducing $Kn > 10$70 to about $Kn > 10$71 m$Kn > 10$72 (Vaidya et al., 2022).

A complementary GOCE case study in the 2026 dissertation uses actual intake areas rather than $Kn > 10$73. With $Kn > 10$74 m$Kn > 10$75, $Kn > 10$76, $Kn > 10$77, and clustered intakes, it reports $Kn > 10$78–$Kn > 10$79 mN and $Kn > 10$80–$Kn > 10$81 mg/s across $Kn > 10$82–$Kn > 10$83 km, with required exhaust velocity $Kn > 10$84–$Kn > 10$85 km/s. Under those assumptions, continuous operation within $Kn > 10$86 kW is feasible above about $Kn > 10$87 km for diffuse or EFD intakes and above about $Kn > 10$88 km for the specular intake, yielding $Kn > 10$89–$Kn > 10$90 mN/kW and $Kn > 10$91–$Kn > 10$92 s (Romano, 2 Jul 2026).

Not all ABEP studies are positive under compact-power assumptions. In the 3U CubeSat particle-based simulation, the intake compresses neutrals in the ionization chamber to $Kn > 10$93 m$Kn > 10$94, $Kn > 10$95 m$Kn > 10$96, and $Kn > 10$97 m$Kn > 10$98, but a power-limited estimate for a $Kn > 10$99 W thruster gives $6{,}000$00–$6{,}000$01 mN, compared with a drag estimate $6{,}000$02 mN for a $6{,}000$03 cm frontal area (Parodi et al., 17 Apr 2025). That study therefore concludes that the modeled CubeSat-scale system cannot station-keep near $6{,}000$04 km unless the frontal area is reduced, the intake capture is increased, the power is raised to the $6{,}000$05–$6{,}000$06 W class, or a lower-$6{,}000$07 operating point is adopted (Parodi et al., 17 Apr 2025). The literature thus separates low-power plasma operation from full system drag compensation.

A qualitatively different path is the cryocondensation-regeneration active intake device, CRAID. Instead of relying only on passive multi-bounce transport, CRAID stores propellant as condensed film on a cryopanel at $6{,}000$08 K, then sublimates it at $6{,}000$09 K and meters it to the thrusters through narrow injection tubes (Moon et al., 3 Mar 2025). At $6{,}000$10 km, the conceptual prototype model gives $6{,}000$11 min, $6{,}000$12 min, $6{,}000$13 min, and $6{,}000$14 min, with effective cycle-averaged $6{,}000$15 and $6{,}000$16 (Moon et al., 3 Mar 2025). The paper states that this compression is at least $6{,}000$17 times higher than that of prevalent passive intake devices and reports complete drag compensation for $6{,}000$18 km under normal solar activity, with no upper altitude bound identified in the $6{,}000$19–$6{,}000$20 km range analyzed (Moon et al., 3 Mar 2025).

Mission applications consequently extend beyond pure stationkeeping. Earth and Mars analyses include circular and elliptical orbit maintenance, orbit raising, constant-rate de-orbit, and atmospheric-propellant collection for refueling or space-tug missions (Vaidya et al., 2022). On Mars, the same study reports full drag compensation near $6{,}000$21 km with $6{,}000$22 and about $6{,}000$23 kW, or near $6{,}000$24 km with $6{,}000$25 and about $6{,}000$26 kW (Vaidya et al., 2022).

6. Ground testing, uncertainties, and open directions

Ground validation is unusually difficult for ABEP because the target environment combines hyperthermal directed flow, extreme rarefaction, and reactive species. The Rarefied Orbital Aerodynamics Research (ROAR) facility is designed around neutral atomic oxygen produced by electron-stimulated desorption, with a baseline DSMC inflow of $6{,}000$27 m/s, $6{,}000$28 K, $6{,}000$29 m$6{,}000$30 s$6{,}000$31, and $6{,}000$32 m$6{,}000$33 (Cushen et al., 2024). Two intake-test methodologies are proposed there: a pressure-difference method using the rise of $6{,}000$34 in an auxiliary chamber, and a gas-sensor method using a quartz crystal microbalance at the outlet (Cushen et al., 2024).

For the pressure-based method, the paper derives

$6{,}000$35

with

$6{,}000$36

so that fitting the transient directly yields $6{,}000$37 once the reverse-performance coefficient $6{,}000$38 is known (Cushen et al., 2024). In the same study, a perfectly specular parabolic inlet gives $6{,}000$39 at $6{,}000$40, while among the tested subscale geometries the Open channel with $6{,}000$41 cm and $6{,}000$42 mm reaches $6{,}000$43 and $6{,}000$44 (Cushen et al., 2024).

A separate numerical analysis studies intake testing with an electric-propulsion plasma plume rather than a neutral AO beam. That hybrid PIC–DSMC model shows that the outlet flow can become neutral-dominated because ions are neutralized on the intake surface, and that sputtering becomes non-negligible at high beam energies. The central experimental conclusion is that simultaneous ion and neutral diagnostics are required for reliable capture-efficiency evaluation when EP plumes are used as a flow generator (Moon et al., 22 Sep 2025). This is an important correction to any testing approach that would infer intake performance from ion current alone.

The principal uncertainties identified across the literature are not merely engineering details. In the learned-kernel work, the molecular-dynamics database is trained primarily for atomic oxygen on amorphous $6{,}000$45; chemistry, adsorption, contamination, and anisotropic roughness are omitted, and temperature dependence at high velocity is approximated through scaling (Schütte et al., 30 Jun 2026). In passive-intake studies, real surfaces are acknowledged to deviate from ideal specular or diffuse behavior under AO exposure, contamination, and microroughness growth (Romano et al., 2021). In coupled neutral–plasma simulations, one-way coupling neglects neutral depletion by ionization and back-pressure from the thruster (Parodi et al., 17 Apr 2025). In active-intake work, CRAID introduces cryocoolers, valves, and mass-flow controllers, so valve reliability, AO-mediated surface chemistry, and long-duration cryosurface contamination become part of the propulsion problem (Moon et al., 3 Mar 2025).

Future directions are correspondingly multi-scale. The gas–surface community names multi-species reactive kernels for $6{,}000$46, $6{,}000$47, and $6{,}000$48, explicit conditioning on wall temperature, material state, and anisotropic roughness, uncertainty quantification, and coupling to plasma solvers and optimization loops (Schütte et al., 30 Jun 2026). System simulations call for two-way neutral–plasma coupling, more complete multi-species chemistry, refined wall models, and integrated power-system modeling (Parodi et al., 17 Apr 2025). Experimental programs prioritize AO wind-tunnel campaigns in ROAR, calibration against ROAR data and SOAR in-orbit measurements, and end-to-end intake–thruster validation under representative rarefied conditions (Romano et al., 2021, Cushen et al., 2024).

Taken together, the current ABEP literature defines a field that is no longer limited to the original passive-intake concept. It now includes high-$6{,}000$49 specular parabolic intakes, diffuse molecular-trap geometries, active cryogenic intakes with cycle-averaged compression ratios near $6{,}000$50, electrodeless RF helicon thrusters with laboratory operation near $6{,}000$51 W, system analyses for Earth and Mars, and thermodynamically constrained learned scattering kernels that close a longstanding gap between molecular dynamics and intake-level rarefied-flow prediction (Romano et al., 2021, Moon et al., 3 Mar 2025, Romano, 2 Jul 2026, Schütte et al., 30 Jun 2026).