The answer in two eccentricities
The paper's headline comparison is between two eccentricity values measured at two different moments in an EMRI's life. An extreme mass ratio inspiral is a stellar-mass compact object spiralling into a massive black hole while emitting gravitational waves. For a massive black hole of 4 × 10^6 solar masses and circular, equal-mass black-hole binaries of 20 solar masses with an initial separation of 0.1 AU, the authors find mean eccentricities of about 0.097 at their adopted LISA-band entry threshold and 0.034 at plunge. The classical channel, driven by two-body relaxation, gives 0.478 and 0.228 at the same two stages.
Those numbers answer the title question in a specific sense: Hills EMRIs are not perfectly circular, but they are systematically rounder than the classical population at both checkpoints. The distinction matters because eccentricity shapes the gravitational-wave signal. More eccentric orbits spread power into higher harmonics and take longer to circularise, while nearly circular orbits produce more monochromatic signals. The paper treats eccentricity as an observable that could help separate formation channels with LISA.
How a binary becomes an EMRI
The Hills mechanism starts with a stellar-mass black-hole binary falling toward a massive black hole on a highly eccentric, nearly parabolic orbit. The massive black hole's tidal forces pull the pair apart: one component is ejected at high speed, and the other is left bound to the massive black hole. If that bound component is a black hole, it can go on to become an EMRI.
What makes this study different is that it follows the whole chain rather than assuming the endpoint. The authors combine an orbit-averaged Fokker–Planck treatment of the nuclear star cluster, three-body modelling of the binary–massive-black-hole encounter, Monte Carlo evolution of the captured black holes, and fully relativistic EMRI trajectories down to final plunge. This is why the paper can connect the rate at which binaries reach the tidal radius to the orbital properties and final fate of the objects they leave behind.
Why Hills orbits stay rounder
The eccentricity difference traces back to where each population decouples from the stellar environment. Classical EMRIs are produced by relaxation-driven diffusion toward the loss cone, so they detach from the cluster relatively close to the separatrix — the last stable orbit — and begin gravitational-wave domination with small semi-latus rectum. The paper reports a mean semi-latus rectum at the start of the gravitational-wave-dominated phase of about 25.5 gravitational radii for classical EMRIs, versus about 83.7 for Hills EMRIs.
Hills-captured black holes, by contrast, are deposited on tighter and more circular orbits than their progenitor binaries and can become gravitational-wave dominated while still far from the loss cone. Because the boundary where relaxation time equals gravitational-wave time is not parallel to the loss-cone boundary, they sit farther from the loss cone when they decouple. They therefore have a longer gravitational-wave-driven phase in which radiation can circularise the orbit before the pericentre shrinks substantially. The authors present this as the mechanism behind the measured distributions, not as a separate measurement.
A broad distribution, not a single value
The mean eccentricity is a summary, and the paper is explicit that the Hills distribution is wide. At LISA-band entry it spans roughly 0.0069 to 0.76, while the classical distribution extends down only to just below the peak of the Hills population. Both distributions share an upper extent near 0.77, which the authors attribute to the maximum eccentricity of a non-plunging orbit at the adopted semi-major axis for this black hole mass.
The tails also correct an earlier expectation. Only 0.09% of Hills EMRIs have an entry eccentricity below 0.01, and 2.83% reach an eccentricity below 0.01 at plunge. That is far fewer than the early analytical estimate that Hills EMRIs would enter the LISA band with almost zero eccentricity. The authors attribute the gap to two things: a more complete treatment of the evolution, which allows black holes to be deposited at smaller pericentres than the analytical estimates assumed, and residual dynamical relaxation that prevents complete circularisation for most systems before they enter the band.
The rate gap behind the shape difference
A cleaner eccentricity distribution does not by itself make Hills EMRIs easy to identify, because they are much rarer in this model. Of the initial black-hole binaries, about 0.39% eventually become EMRIs, whereas about 59% of captured black holes do. The bottleneck is getting the binary into the tidal sphere, not converting a captured black hole into an EMRI.
For comparable initial compact-object mass in the nuclear star cluster, the paper finds about 1.6 × 10^-8 Hills EMRIs per year per galaxy, roughly 12% of the classical yield — about 88% fewer events. Concretely, 93 realisations containing 40,000 binaries each yield 14,517 Hills EMRIs, or 156 per galaxy over 10 Gyr, against 1,344 classical EMRIs per galaxy over the same period. These comparisons assume the specific nuclear star cluster model and binary properties adopted in the paper.
What a low eccentricity measurement can tell
The paper turns the two distributions into a conditional probability: given a measured eccentricity, how likely is a Hills origin? The answer depends strongly on μ, the fraction of compact-object mass initially in binaries, which the authors treat as a free parameter. Without rate weighting, EMRIs with plunge eccentricity below about 0.087 are more likely than not to be Hills events under the reference model. Once relative rates and μ are included, that association weakens sharply.
For μ = 0.1, the probability of a Hills origin stays below 20% even for the most circular systems at plunge; for μ = 0.01 it stays below 5% across the eccentricity range. At LISA-band entry the separation between the distributions is larger, so low eccentricity is more informative: for μ = 1/2 the conditional probability reaches 50% at about e_LISA ≤ 0.12 and stays above 20% for e_LISA ≤ 0.25, while for μ = 1/10 the low-eccentricity region carries 5% to 20% probability up to e_LISA ≈ 0.20. The authors stress that these are population-level probabilities under their assumed cluster model, binary fraction and rates, and that they do not include LISA selection effects.
Where the conclusion stops
The quantitative boundaries in this study are tied to one configuration: a 4 × 10^6 solar-mass black hole, equal-mass circular binaries of 20 solar masses, and a 0.1 AU initial separation. The authors state that the encounter outcomes depend on black hole mass and binary properties, and that their quantitative boundaries should not be regarded as universal. The single black hole mass also sits toward the high-mass end of the LISA detection range, while detections are likely to be dominated by systems with masses below about 10^6 solar masses; assessing how well the two channels can be separated statistically will require extending the analysis to lower masses.
The quasi-periodic eruption discussion carries a similar boundary. The paper does not establish the Hills mechanism as a general explanation for those transients. Instead it offers a physically motivated diagnostic to apply once more reliable, model-independent constraints on their orbital parameters become available.
The priority the binary fraction sets
The most consequential variable in this paper is not the eccentricity measurement itself but the binary mass fraction in nuclear star clusters. The shape difference between Hills and classical EMRIs is robust in the authors' framing: Hills systems are rounder at both plunge and LISA-band entry. What changes the interpretation is how many binaries the cluster actually holds. If binaries are abundant, a very circular EMRI is a meaningful hint of a Hills origin; if they are rare, even the roundest systems remain more likely to be classical.
That makes the practical priority clear for anyone planning to use LISA eccentricity measurements to identify formation channels: pair the measured eccentricity with a population model for the binary fraction, and read the result at LISA-band entry rather than at plunge, where the two distributions overlap more. The paper's own numbers put the informative regime at entry eccentricities below roughly 0.12 to 0.25 for a binary fraction of one-half, and much weaker for lower fractions.
Sources used in this answer
Extreme mass ratio inspirals from the Hills mechanism, The eccentricity distribution of Hills EMRIs in LISA
Self-consistent modelling shows Hills EMRIs enter the LISA band with mean eccentricity about 0.097 and plunge at about 0.034, versus 0.478 and 0.228 for classical EMRIs, while producing roughly 12% of the classical EMRI yield for comparable initial compact-object mass.
