NRSurE_q4NoSpin_22: Eccentric BBH Surrogate

• The paper introduces NRSurE_q4NoSpin_22, a full inspiral-merger-ringdown surrogate that models the (2,2) gravitational-wave mode for eccentric, nonspinning binary black holes using a novel radial-phase-domain approach.
• The model is trained on 156 numerical-relativity waveforms from the SXS catalog, covering mass ratios in [1,4] and eccentricities from 10⁻³ to 0.25, achieving median mismatches of 2×10⁻⁵.
• By blending a radial-phase inspiral surrogate with a time-domain merger surrogate, the model overcomes eccentric dephasing challenges to enable high-accuracy gravitational-wave parameter estimation.

Searching arXiv for the cited waveform and companion surrogate papers to ground the article with current metadata. arxiv_search.query({"search_query":"id:(Nee et al., 30 Sep 2025) OR id:(Ravichandran et al., 30 Apr 2026)","start":0,"max_results":10}) arxiv_search.query returned 2 results:

1. (Nee et al., 30 Sep 2025) — "Eccentric binary black holes: A new framework for numerical relativity waveform surrogates"
2. (Ravichandran et al., 30 Apr 2026) — "Merger remnant and eccentricity dynamics surrogates for eccentric nonspinning black hole binaries" NRSurE_q4NoSpin_22 is a numerical-relativity surrogate model for eccentric, non-spinning binary black hole mergers, constructed specifically for the dominant $(2,2)$ spin-weighted spherical-harmonic mode of the gravitational-wave signal. It is a full inspiral-merger-ringdown model, but its construction departs from standard quasi-circular surrogate practice: the inspiral is represented in a radial-phase domain, while the merger-ringdown is modeled in the time domain and then stitched to the inspiral piece. The model is trained on numerical-relativity waveforms from the SXS catalog over the parameter domain $q\in[1,4]$, $e_{-3000M}\in[10^{-3},0.25]$, and $\ell_{-1200M}\in[0,2\pi]$ for production use, where $t=0$ is measured relative to merger and $M$ is the total mass (Nee et al., 30 Sep 2025). A closely related later paper introduces companion dynamics and remnant surrogates whose parameterization is chosen to match this waveform model, which situates NRSurE_q4NoSpin_22 within a broader eccentric, nonspinning surrogate family (Ravichandran et al., 30 Apr 2026).

1. Definition, nomenclature, and modeled systems

The name of the model is literal. “NRSur” denotes a numerical-relativity surrogate, “E” denotes eccentric, “q4” indicates training up to mass ratio $q\leq 4$, “NoSpin” restricts the model to non-spinning binaries, and “22” indicates that only the $(\ell,m)=(2,2)$ mode is included (Nee et al., 30 Sep 2025). No higher harmonics are part of NRSurE_q4NoSpin_22 as presented.

The model begins from the complex strain

$h=h_+ - i h_\times,$

decomposes it into spin-weighted spherical-harmonic modes, and retains only

$h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$

Its reference time convention sets

$q\in[1,4]$0

at the peak of

$q\in[1,4]$1

The surrogate is therefore defined in terms of the $q\in[1,4]$2-mode amplitude $q\in[1,4]$3 and phase $q\in[1,4]$4, rather than a multimode waveform representation.

The covered systems are non-spinning eccentric binary black holes. The training domain is reported as

$q\in[1,4]$5

with the mean-anomaly parameter $q\in[1,4]$6 defined at $q\in[1,4]$7 before merger. The waveform length is not fixed, because the inspiral is built in radial phase rather than on a fixed time grid. The reported duration is approximately $q\in[1,4]$8, ranging from $q\in[1,4]$9 to $e_{-3000M}\in[10^{-3},0.25]$0, corresponding to about 42 gravitational-wave cycles of the $e_{-3000M}\in[10^{-3},0.25]$1 mode (Nee et al., 30 Sep 2025).

2. Motivation and the eccentric-surrogate problem

The model was developed to address a specific failure mode of standard surrogate methods when applied to eccentric binaries. For quasi-circular, non-precessing systems, one often models $e_{-3000M}\in[10^{-3},0.25]$2 and $e_{-3000M}\in[10^{-3},0.25]$3 directly in time. For eccentric binaries, however, eccentricity introduces oscillations in amplitude and phase on the orbital or radial timescale, explicit dependence on the mean anomaly $e_{-3000M}\in[10^{-3},0.25]$4, and parameter-dependent variation of the radial frequency. As a result, waveforms that are initially aligned in time move in and out of phase across parameter space, making the data poorly compressible and difficult to fit with reduced bases and interpolation (Nee et al., 30 Sep 2025).

The paper identifies two broad difficulties in eccentric numerical-relativity surrogates. The first is covering parameter space with numerical-relativity simulations, because eccentric binaries are sensitive to initial conditions. The second is representational: even with simulations in hand, the standard quasi-circular decomposition is not effective. The central challenge is temporal incoherence of eccentric features across parameter space. Unlike the precession problem, for which co-precessing-frame constructions are available, there is no analogous frame transformation that removes eccentricity-induced oscillations. The reported result is that “modern techniques for quasi-circular systems” give mismatches several orders of magnitude worse than the eccentric decomposition used in NRSurE_q4NoSpin_22 (Nee et al., 30 Sep 2025).

This makes the model notable less for extending a standard pipeline than for reformulating the representation problem itself. A plausible implication is that its scientific significance lies as much in the eccentric decomposition as in the fitted surrogate.

3. Radial-phase-domain inspiral representation

The defining methodological innovation is the inspiral surrogate in radial phase. Instead of fitting the inspiral directly as functions of time, the model extracts a radial phase $e_{-3000M}\in[10^{-3},0.25]$5 and represents

$e_{-3000M}\in[10^{-3},0.25]$6

This factorization is the core of the construction (Nee et al., 30 Sep 2025).

The chosen radial phase is the relativistic anomaly from quasi-Keplerian post-Newtonian equations. The paper quotes a system for $e_{-3000M}\in[10^{-3},0.25]$7, $e_{-3000M}\in[10^{-3},0.25]$8, and

$e_{-3000M}\in[10^{-3},0.25]$9

including 3PN terms, with $\ell_{-1200M}\in[0,2\pi]$0 the symmetric mass ratio, $\ell_{-1200M}\in[0,2\pi]$1 the total mass, and $\ell_{-1200M}\in[0,2\pi]$2 the instantaneous orbital frequency. The relativistic anomaly is selected because, unlike mean anomaly used in some previous work, it is described as a smooth function of time with a well-defined $\ell_{-1200M}\in[0,2\pi]$3 limit (Nee et al., 30 Sep 2025).

To obtain $\ell_{-1200M}\in[0,2\pi]$4, the construction supplies

$\ell_{-1200M}\in[0,2\pi]$5

with $\ell_{-1200M}\in[0,2\pi]$6 in the non-spinning model. The mass ratio $\ell_{-1200M}\in[0,2\pi]$7 is taken from Christodoulou masses from apparent horizons, $\ell_{-1200M}\in[0,2\pi]$8 is computed from coordinate centers of the black holes, and $\ell_{-1200M}\in[0,2\pi]$9 are estimated using waveform-based eccentricity tools. The paper exploits that at periastron,

$t=0$0

and extracts a waveform-based eccentricity $t=0$1 at the first available periastron using gw_eccentricity; although $t=0$2, it is found close enough for the surrogate construction (Nee et al., 30 Sep 2025).

The practical effect of the change of variables is to phase-align the eccentric oscillatory structure. In $t=0$3 and $t=0$4, the eccentric oscillations are mapped to the same period and are in phase across parameter space, while variation of the radial or orbital timescale is absorbed into the monotonic map $t=0$5. This is how the model factors out eccentricity-induced dephasing. During preprocessing, the inspiral surrogate keeps the 11 radial periods immediately before the first periastron passage after $t=0$6, and each waveform is shifted and rotated so that at the final modeled periastron,

$t=0$7

A further complication is that the inspiral data pieces are not strictly periodic in $t=0$8, even though systems differing by $t=0$9 in mean anomaly are physically equivalent. The paper describes this as pseudo-periodicity: shifting $M$0 by $M$1 corresponds to the same physical state shifted by one radial period, so the waveform segment is shifted rather than pointwise identical. To improve fits near $M$2 and $M$3, the training set is augmented with copies shifted forward and backward by one radial period, giving an effective training domain

$M$4

while production use remains

$M$5

For the merger surrogate, by contrast, exact duplication with $M$6 is sufficient (Nee et al., 30 Sep 2025).

4. Time-domain merger-ringdown surrogate and stitching

Near merger, the radial-phase construction ceases to be reliable. The stated reasons are that black-hole positions cease to be tracked and the post-Newtonian-based $M$7 extraction becomes unreliable in the strong-field regime. NRSurE_q4NoSpin_22 therefore introduces a second surrogate, built directly in the time domain for the final $M$8 orbits plus merger and ringdown (Nee et al., 30 Sep 2025).

This merger surrogate models

$M$9

on the interval

$q\leq 4$0

It uses the same regression coordinates as the inspiral surrogate: $q\leq 4$1 The paper notes that this representation is less compressible than the radial-phase inspiral representation, but the segment is short enough that a reduced basis still works well and there is limited opportunity for large dephasing over only $q\leq 4$2 orbits (Nee et al., 30 Sep 2025).

The full inspiral-merger-ringdown waveform is assembled by evaluating both surrogates, aligning them in time and phase, and smoothly blending them. Let $q\leq 4$3 denote the inspiral surrogate’s internal time coordinate. Since the inspiral piece is shifted so that its final modeled periastron is at $q\leq 4$4, the location of $q\leq 4$5 in this coordinate must be inferred from $q\leq 4$6. The paper gives

$q\leq 4$7

followed by the time shift

$q\leq 4$8

After this, both waveforms are rotated so that

$q\leq 4$9

and then stitched with Planck window functions centered at

$(\ell,m)=(2,2)$0

with width

$(\ell,m)=(2,2)$1

The resulting construction is therefore a hybrid of an inspiral surrogate in $(\ell,m)=(2,2)$2 and a merger surrogate in $(\ell,m)=(2,2)$3, joined through a parameter-dependent alignment inferred from mean anomaly (Nee et al., 30 Sep 2025).

5. Training set, preprocessing, and validation

The model is trained on 156 numerical-relativity waveforms from the SXS catalog: $(\ell,m)=(2,2)$4 The preprocessing pipeline consists of mode extraction and amplitude-phase decomposition, definition of the peak-amplitude reference time, radial-phase extraction, inspiral truncation to 11 radial periods before the first periastron after $(\ell,m)=(2,2)$5, inspiral alignment, construction of the merger segment on $(\ell,m)=(2,2)$6, duplication in $(\ell,m)=(2,2)$7 to enforce periodicity or pseudo-periodicity, and Gaussian process regression across parameter space (Nee et al., 30 Sep 2025).

Validation uses leave-eight-out cross validation: the 156 waveforms are randomly divided into groups of 8, a surrogate is built omitting one group, validated on the excluded waveforms, and the process is repeated over groups. The headline performance numbers reported in the abstract are a maximum mismatch of

$(\ell,m)=(2,2)$8

and a median mismatch of

$(\ell,m)=(2,2)$9

The paper also compares surrogate mismatches to mismatches between the two highest numerical-relativity resolutions for each system, using those as a conservative estimate of waveform error. Its conclusion is that the surrogate reproduces the numerical-relativity waveforms to within their intrinsic errors, and that the limiting factor is likely numerical-relativity accuracy rather than surrogate fitting accuracy (Nee et al., 30 Sep 2025).

The worst validation case is reported to be one of the largest-$h=h_+ - i h_\times,$0, largest-$h=h_+ - i h_\times,$1 systems, near the edge of the training domain. The paper states that adding simulations at larger eccentricity and mass ratio would likely improve performance there. It also emphasizes that no formal extrapolation study is given and that the model should not be assumed reliable outside

$h=h_+ - i h_\times,$2

This boundary sensitivity is consistent with the general surrogate-model caveat that calibration domain, rather than architecture alone, determines reliable use.

6. Practical use, interpretation, and limitations

NRSurE_q4NoSpin_22 is positioned for high-accuracy parameter estimation for eccentric binary black holes, astrophysical inference on formation channels, tests of general relativity, and as a template or benchmark for accelerating other eccentric waveform models (Nee et al., 30 Sep 2025). Because it is a numerical-relativity surrogate, its practical role is to deliver fast evaluation with high faithfulness to full numerical relativity inside a controlled interpolation domain.

Several limitations are explicit. The model includes only the $h=h_+ - i h_\times,$3 mode; it is non-spinning only; it is restricted to $h=h_+ - i h_\times,$4 and $h=h_+ - i h_\times,$5; its length is only about $h=h_+ - i h_\times,$6–$h=h_+ - i h_\times,$7, roughly 42 gravitational-wave cycles; and there is no guarantee outside the training domain (Nee et al., 30 Sep 2025). The radial-phase extraction also relies internally on post-Newtonian equations and coordinate trajectory information.

The implementation note given in the paper is that NRSurE_q4NoSpin_22 “will be made publicly available as part of the GWSurrogate package” (Nee et al., 30 Sep 2025). This indicates intended public deployment in the same software ecosystem used for related waveform and dynamics surrogates.

A common misunderstanding is to treat the model as a generic eccentric waveform approximant. Its calibrated status is narrower: it is a specific numerical-relativity surrogate for non-spinning systems and for the dominant mode only. Another common misunderstanding is to infer multimode content from the full inspiral-merger-ringdown scope. The “22” in the name is literal: the surrogate models only $h=h_+ - i h_\times,$8.

7. Position within the NRSurE_q4NoSpin family

A later companion paper introduces two additional eccentric, nonspinning surrogates: a remnant model for $h=h_+ - i h_\times,$9 and a dynamics model for $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$0 and $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$1. That paper is directly relevant because it states explicitly that the dynamics surrogate’s parameterization was chosen to match NRSurE_q4NoSpin_22 (Ravichandran et al., 30 Apr 2026). In that shared family-level convention, the dynamics surrogate uses

$h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$2

and the time interval

$h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$3

whereas the remnant surrogate uses

$h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$4

This relation clarifies the status of NRSurE_q4NoSpin_22 within the broader eccentric-surrogate ecosystem. The waveform model and the dynamics surrogate share reference-time conventions, while the remnant surrogate uses a later reference time because remnant quantities fit better there. The companion paper further states that the dynamics surrogate can map between these parameterizations: starting from waveform-style parameters $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$5, one predicts $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$6 and $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$7, reads off $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$8, and then evaluates the remnant surrogate (Ravichandran et al., 30 Apr 2026). The same paper reports that doing this introduces only errors comparable to surrogate and numerical-relativity resolution errors.

The companion paper is also explicit about what it does not provide for NRSurE_q4NoSpin_22: it does not define or construct the waveform model itself, does not give its surrogate architecture, and does not report its waveform accuracy metrics (Ravichandran et al., 30 Apr 2026). That distinction matters bibliographically. The waveform model’s own definition, decomposition, domain, and mismatch results belong to the waveform paper, while the later work supplies context on family-wide parameter conventions, dynamics, and remnant inference. This suggests that NRSurE_q4NoSpin_22 is best understood not as an isolated surrogate, but as the waveform component of a coordinated eccentric, nonspinning $h_{22}(t)=A_{22}(t)e^{-i\phi_{22}(t)}.$9 surrogate framework.