Reliable equation-of-state constraints from binary-neutron-star post-merger gravitational waves: out-of-sample model systematics, training-data limits, and exotic-matter robustness

When two neutron stars merge, the hot, differentially rotating remnant emits gravitational waves at kilohertz frequencies that encode the behavior of matter at supranuclear densities — information inaccessible from the inspiral alone. Easter et al. (2020) built a damped-sinusoid post-merger waveform model and a hierarchical inference model, trained on 35 numerical-relativity simulations, that constrains the progenitor tidal coupling constant and compactness; they showed the pipeline works on injections but identified three unresolved obstacles to reliable equation-of-state constraints from real detections: systematic out-of-sample bias of the inference model when training and target simulations come from different numerical-relativity codes, resolutions, and spectral structures; the small size and limited coverage of the training set; and untested robustness against post-merger signals that deviate quantitatively from current simulations, for instance due to a first-order phase transition to deconfined quark (strange) matter. The 2020–2026 follow-up literature includes NR-informed models trained on hundreds of simulations, supervised-learning and neural-network surrogates, full-spectrum Bayesian inference for next-generation detectors, and sub-threshold population stacking; while training-data scale and exotic-matter content of models have grown dramatically, no published analysis demonstrates unbiased out-of-sample recovery of tidal, compactness, or equation-of-state-level quantities with calibrated credible intervals — the core problem remains open.

Background

A binary neutron star (BNS) coalescence has three gravitational-wave (GW) stages: a long inspiral, observable by current detectors such as Advanced LIGO and Virgo; a merger; and a post-merger stage in which the remnant — a hot, massive neutron star or, if it is too massive, a promptly forming black hole — rings at characteristic frequencies of roughly 2–4 kHz. The equation of state (EOS), i.e. the relation between density and pressure of cold matter above nuclear saturation density, is only weakly constrained by the inspiral (through the tidal deformability, the induced quadrupole of each star in the companion's tidal field). The post-merger spectrum, by contrast, probes hot matter at even higher densities and can reveal phase transitions to deconfined quark matter that leave the cold inspiral signal nearly unchanged. No post-merger signal has ever been detected: the search following GW170817, the first BNS detection, returned only upper limits, because current detectors are barely sensitive at kilohertz frequencies. Next-generation detectors — the Einstein Telescope and Cosmic Explorer, plus proposed high-frequency detectors — are expected to change this, which makes trustworthy post-merger analysis a gating capability for dense-matter physics.

The key theoretical input is a set of quasi-universal relations found in numerical-relativity (NR) simulations, which solve Einstein's equations coupled to fluid dynamics at a cost of order $10^5$ CPU-hours per simulation. In particular, the dominant post-merger spectral peak $f_{\mathrm{peak}}$ correlates tightly with the tidal coupling constant $\kappa_{2}^{\mathrm{T}}$ — a mass-weighted combination of the stars' tidal deformabilities — and with the stellar compactness $C$ (the ratio of mass to radius in gravitational units). Rezzolla and Takami (2016) systematically mapped these spectral relations over a grid of simulations with varying equations of state and masses. Easter et al. (2019) exploited this to build a hierarchical model: a linear model, trained on 35 Rezzolla–Takami spectra, that predicts the full post-merger amplitude spectrum from the progenitor mass and $\kappa_{2}^{\mathrm{T}}$, so that an observed spectrum can be inverted to infer $\kappa_{2}^{\mathrm{T}}$ and $C$. Because NR simulations are too expensive to use as templates directly, such trained surrogate models are the only practical route to matched filtering (cross-correlating the data with a waveform template to extract a known signal from noise) and to Bayesian parameter estimation (the standard statistical framework that returns posterior probability distributions for source parameters) of post-merger signals. The accuracy of a template is quantified by the fitting factor, the maximal matched-filter overlap with the true signal (values above about 0.97 lose less than 10% of the detection rate).

Easter et al. (2020) then performed the first end-to-end test: a time-domain analytic model of three exponentially damped sinusoids with linear frequency drift, a Bayes-factor detection statistic (the Bayesian odds for signal versus noise), and the 2019 hierarchical model for EOS inference, validated on NR waveforms from the CoRe database — an independent multi-code simulation catalog. The test exposed the problem at issue. Applied out-of-sample (trained on Rezzolla–Takami simulations, tested on CoRe simulations), the hierarchical model's posteriors were biased: the injected compactness fell outside the 95% credible interval. The authors attributed the bias to three factors — the training and injection simulations differ in numerical codes, resolution, and spectral structure; 35 training waveforms is a minimal set; and NR waveforms themselves depend on resolution (e.g. collapse time shifts). They further noted that a first-order phase transition in the hot remnant (as studied by Bauswein et al. 2019) may make a real post-merger signal quantitatively different from any current simulation, and that their model's robustness to such deviations — the very scenario that makes post-merger observations scientifically most interesting — was untested. The unresolved objective is therefore: post-merger detection and inference whose parameter and EOS-level claims survive contact with signals from simulations, and ultimately a real event, that the model was never trained on.

Problem Statement

The open problem is to make detection of, and parameter inference from, binary-neutron-star post-merger gravitational-wave signals reliable enough to support equation-of-state constraints — that is, to develop post-merger waveform and hierarchical inference frameworks mapping the post-merger signal to progenitor and dense-matter properties (the tidal coupling constant $\kappa_{2}^{\mathrm{T}}$, the compactness $C$, and EOS-level quantities such as the maximum density or maximum mass) such that:

(i) the inference is unbiased out-of-sample: applied to waveforms from numerical-relativity simulations not used in training — from independent codes, at different resolutions, and with different spectral structure — it recovers the injected $\kappa_{2}^{\mathrm{T}}$, $C$, and EOS-level quantities with any residual bias quantified and subdominant to the statistical uncertainty, and with correct credible-interval coverage whose intervals are informative — narrow enough to exclude the uninformative prior-only posterior and to constrain $\kappa_{2}^{\mathrm{T}}$, $C$, and the EOS-level quantities well below the prior range — and the training-set size and coverage (in mass, mass ratio, and equation-of-state family) required to achieve this are characterized;

(ii) the analysis is robust against — that is, retains the out-of-sample unbiasedness, credible-interval coverage, and informativeness required by (i) for — post-merger signals that deviate quantitatively from current numerical-relativity predictions, in particular a strong first-order phase transition to deconfined quark (strange) matter or other exotic physics in the hot remnant;

(iii) these properties hold at the post-merger signal-to-noise ratios realistically expected for detectable events (order 7–50, the detection-threshold and precision-forecast regime of the source paper), so that equation-of-state constraints derived from an actual post-merger detection are trustworthy rather than systematics-dominated.

Conditions (i)–(iii) are jointly required to resolve the task: the framework must satisfy all three, not a proper subset.

The scientific objective uniting all components is that systematic uncertainties, not statistical noise, no longer dominate EOS-level claims from post-merger observations.

References

1. Paul J. Easter, Sudarshan Ghonge, Paul D. Lasky, Andrew R. Casey, James A. Clark, Francisco Hernandez Vivanco, Katerina Chatziioannou, 'Detection and parameter estimation of binary neutron star merger remnants', Physical Review D 102, 043011 (2020). DOI 10.1103/PhysRevD.102.043011; arXiv:2006.04396; https://arxiv.org/abs/2006.04396
2. Paul J. Easter, Paul D. Lasky, Andrew R. Casey, Luciano Rezzolla, Kentaro Takami, 'Computing fast and reliable gravitational waveforms of binary neutron star merger remnants', Physical Review D 100, 043005 (2019). DOI 10.1103/PhysRevD.100.043005; arXiv:1811.11183; https://arxiv.org/abs/1811.11183
3. Luciano Rezzolla, Kentaro Takami, 'Gravitational-wave signal from binary neutron stars: a systematic analysis of the spectral properties', Physical Review D 93, 124051 (2016). DOI 10.1103/PhysRevD.93.124051; arXiv:1604.00246; https://arxiv.org/abs/1604.00246
4. Andreas Bauswein, Niels-Uwe F. Bastian, David B. Blaschke, Katerina Chatziioannou, James A. Clark, Tobias Fischer, Micaela Oertel, 'Identifying a first-order phase transition in neutron-star mergers through gravitational waves', Physical Review Letters 122, 061102 (2019). DOI 10.1103/PhysRevLett.122.061102; arXiv:1809.01116; https://arxiv.org/abs/1809.01116
5. B. P. Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), 'Search for post-merger gravitational waves from the remnant of the binary neutron star merger GW170817', Astrophysical Journal Letters 851, L16 (2017). DOI 10.3847/2041-8213/aa9a35; arXiv:1710.09320; https://arxiv.org/abs/1710.09320
6. Matteo Breschi, Sebastiano Bernuzzi, Kabir Chakravarti, Alessandro Camilletti, Aviral Prakash, Albino Perego, 'Kilohertz gravitational waves from binary neutron star mergers: Numerical-relativity informed postmerger model' (NRPMw), Physical Review D 109, 064009 (2024). DOI 10.1103/PhysRevD.109.064009; arXiv:2205.09112; https://arxiv.org/abs/2205.09112
7. Matteo Breschi, Rossella Gamba, Gregorio Carullo, Ssohrab Borhanian, Sebastiano Bernuzzi, 'Kilohertz gravitational waves from binary neutron star mergers. II. Inference of postmerger signals with the Einstein Telescope', Physical Review D 112, 124001 (2025). DOI 10.1103/nmns-md9c; arXiv:2205.09979; https://arxiv.org/abs/2205.09979
8. Giulia Huez, Sebastiano Bernuzzi, Matteo Breschi, Rossella Gamba, 'Kilohertz gravitational waves from binary neutron star mergers: Full spectrum analyses and high-density constraints on neutron star matter', Physical Review D 113, 043019 (2026). DOI 10.1103/2jxk-1bf1; arXiv:2507.06293; https://arxiv.org/abs/2507.06293
9. Theodoros Soultanis, Kiril Maltsev, Andreas Bauswein, Katerina Chatziioannou, Friedrich K. Roepke, Nikolaos Stergioulas, 'Gravitational-wave model for neutron star merger remnants with supervised learning', Physical Review D 111, 023002 (2025). DOI 10.1103/PhysRevD.111.023002; arXiv:2405.09513; https://arxiv.org/abs/2405.09513
10. Theodoros Soultanis, Andreas Bauswein, Nikolaos Stergioulas, 'Analytic models of the spectral properties of gravitational waves from neutron star merger remnants', Physical Review D 105, 043020 (2022). DOI 10.1103/PhysRevD.105.043020; arXiv:2111.08353; https://arxiv.org/abs/2111.08353
11. Alejandra Gonzalez, Francesco Zappa, Matteo Breschi, Sebastiano Bernuzzi, David Radice, Ananya Adhikari, Alessandro Camilletti, Swami Vivekanandji Chaurasia, Georgios Doulis, Surendra Padamata, Alireza Rashti, Maximiliano Ujevic, Bernd Bruegmann, William Cook, Tim Dietrich, Albino Perego, Amit Poudel, Wolfgang Tichy (CoRe collaboration), 'Second release of the CoRe database of binary neutron star merger waveforms', Classical and Quantum Gravity 40, 085011 (2023). DOI 10.1088/1361-6382/acc231; arXiv:2210.16366; https://arxiv.org/abs/2210.16366
12. Aviral Prakash, Ish Gupta, Matteo Breschi, Rahul Kashyap, David Radice, Sebastiano Bernuzzi, Domenico Logoteta, B. S. Sathyaprakash, 'Detectability of QCD phase transitions in binary neutron star mergers: Bayesian inference with the next generation gravitational wave detectors', Physical Review D 109, 103008 (2024). DOI 10.1103/PhysRevD.109.103008; arXiv:2310.06025; https://arxiv.org/abs/2310.06025
13. Stamatis Vretinaris, Georgios Vretinaris, Christos Mermigkas, Minas Karamanis, Nikolaos Stergioulas, 'Robust and fast parameter estimation for gravitational waves from binary neutron star merger remnants', Physical Review D 113, 024012 (2026). DOI 10.1103/g1qs-j74x; arXiv:2501.11518; https://arxiv.org/abs/2501.11518
14. Dimitrios Pesios, Nikolaos Stergioulas, 'Inferring neutron-star properties from post-merger gravitational-wave spectra with neural networks', arXiv preprint (2026). arXiv:2605.24145; https://arxiv.org/abs/2605.24145
15. Fiona H. Panther, Paul D. Lasky, 'Sub-threshold post-merger gravitational waves can constrain the hot nuclear equation of state', Monthly Notices of the Royal Astronomical Society 548, stag750 (2026). DOI 10.1093/mnras/stag750; arXiv:2512.08497; https://arxiv.org/abs/2512.08497
16. Matteo Breschi, Sebastiano Bernuzzi, Daniel Godzieba, Albino Perego, David Radice, 'Constraints on the maximum densities of neutron stars from postmerger gravitational waves with third-generation observations', Physical Review Letters 128, 161102 (2022). DOI 10.1103/PhysRevLett.128.161102; arXiv:2110.06957; https://arxiv.org/abs/2110.06957
17. Sebastian Blacker, Hristijan Kochankovski, Andreas Bauswein, Angels Ramos, Laura Tolos, 'Thermal behavior as indicator for hyperons in binary neutron star merger remnants', Physical Review D 109, 043015 (2024). DOI 10.1103/PhysRevD.109.043015; arXiv:2307.03710; https://arxiv.org/abs/2307.03710