WorksPhysics and Astronomy
The MARTINI platform II. Tracing r-process nucleosynthesis in neutron star mergers with long-lived remnants
Abstract
Binary neutron star (BNS) mergers are among the primary sites of the rapid neutron-capture (r-)process. In mergers with long-lived remnants, sustained neutrino irradiation raises the electron fraction of the disc wind ejecta, modifying the resulting nucleosynthesis pattern. Yields from numerical-relativity simulations with accurate neutrino transport are needed to quantify the contribution of BNS mergers to the Galactic production of heavy elements and to interpret kilonova spectra. We present and publicly release tracer trajectories and nucleosynthesis yields for eight BNS mergers forming long-lived neutron star remnants, obtained from general-relativistic simulations with M1 neutrino transport, spanning two equations of state (DD2 and SFHo), and a range of total masses and mass ratios. Passive tracer particles following the dynamical and wind ejecta were post-processed with the nuclear reaction network WinNet up to 1 Gyr. The yields and tracer trajectories are available on the MARTINI platform. At 1 day, the production of light and first-peak elements (He, Ni, Sr, and Y) is similar across models and is primarily enhanced by the neutrino-irradiated disc wind. The yields of heavy nuclei vary by more than one order of magnitude, with unequal-mass binaries producing the largest amounts owing to their more neutron-rich ejecta. Our He yields, particularly in the wind, exceed those of recent studies based on an M0 neutrino scheme, consistent with the stronger neutrino irradiation resulting from an M1 scheme. At 1 Gyr, all models reproduce the second and third solar r-process peaks, but differ in the first peak, the lanthanide region, the mass range between the first and second peaks, and the trans-lead region, reflecting the electron fraction distribution of the ejecta and the dynamical and wind contributions. Two equal-mass models provide a close match to the solar pattern.
Introduction
The astrophysical origin of elements heavier than iron (Fe) remains one of the longest-standing problems in nuclear astrophysics [69]. Approximately half of these nuclei are synthesised through the rapid neutron-capture (-) process [11], which is responsible for the production of the heaviest stable elements in nature, including the actinides , , and (for recent reviews on -process, see Thielemann et al. [114]; Cowan et al. [15]). Because the -process requires extremely neutron-rich conditions, with neutron densities exceeding to produce elements up to actinides, and a short timescale ( s), identifying the astrophysical environments in which it operates has been a major challenge for decades. Several astrophysical sites have been proposed as potential sources of the -process, including core-collapse supernovae, magneto-rotational explosions, collapsars, and compact-object mergers. While some
of these environments may contribute to a weak -process [117, 43], growing theoretical and observational evidence indicates that binary neutron star (BNS) mergers, and possibly neutron star (NS)-black hole (BH) mergers, are among the primary production sites of the main -process responsible for the Solar abundance pattern (e.g. [107, 101, 44, 103].
Compelling evidence for this scenario came with the epochal discovery of GW170817, the first gravitational-wave detection of a BNS merger, and the observation of its associated UV/optical/NIR counterpart, the kilonova AT2017gfo [1 b]. AT2017gfo exhibited clear signatures of freshly synthesised r-process material, providing the first direct observational confirmation that BNS mergers produce heavy elements through -process (e.g. [79, 106]). Additional observational evidence has further strengthened interest in understanding -process nucleosynthesis in the ejecta of compact binary mergers. In particular, the abundance patterns observed in extremely metal-poor stars, including the so-called actinide-boost stars characterised by enhanced thorium (Th) and uranium (U) abundances relative to lighter -process elements such as europium (Eu), provide a unique probe of the astrophysical conditions under which the heaviest nuclei are synthesised and raise important questions about the diversity of -process sites (e.g. [41, 42, 24, 57] (Ji & Frebel 2018)). Eichler et al. [24] showed that the Th/Eu ratio is highly sensitive to both the thermodynamic conditions of the ejecta and the adopted nuclear input, particularly the nuclear mass model, suggesting that the observed diversity of Th/Eu ratios may arise either from different ejecta components within BNS mergers or from multiple astrophysical -process sites.
In this context, accurate nucleosynthesis predictions from BNS mergers are essential not only to quantify the contribution of these events to the Galactic production of heavy elements, but also to interpret kilonova observations and to assess whether BNS mergers can reproduce the chemical signatures observed in metal-poor and actinide-boost stars.
The nucleosynthesis outcome from a BNS merger depends strongly on the properties of the ejected matter, in particular its electron fraction, entropy, and expansion velocity ([39 b]). These properties depend sensitively on several intrinsic binary properties, such as the total mass, mass ratio and spins. Additionally, they vary subject to several microphysics inputs, including the nuclear equation of state (EOS) and the treatment of weak interactions, requiring fully general-relativistic simulations coupled to neutrino-radiation hydrodynamics (see [86, 84, 77] for recent reviews). In particular, the lifetime of the merger remnant plays a crucial role in shaping the properties of the remnant accretion disc and the subsequent outflows. Long-lived NS remnants sustain neutrino irradiation over timescales comparable to the disc lifetime, increasing the electron fraction of disc wind ejecta and substantially modifying the resulting nucleosynthesis pattern (e.g. [31, 51, 50, 8, 109]). The predicted ejecta composition is also highly sensitive to the adopted neutrino transport scheme. Specifically, the M1 neutrino transport scheme ([28, 85, 122]) consistently predicts broader electron fraction distributions in the neutrino-driven wind than the simpler M0 approximation, leading to enhanced production of light and Fe-peak nuclei (e.g. [28, 116, 85, 82, 83, 8, 47, 109]). The angular structure of the neutrino irradiation in M1 simulations and its imprint on the electron fraction was recently quantified by Çokluk et al. [12] for the set of BNS merger simulations considered in this work. They found that the quadrupole-dominated neutrino radiation field drives the polar material close to the equilibrium electron fraction set by neutrino absorption, with values of about 0.50–0.55 at late times, whereas the equatorial material remains more neutron-rich and below its equilibrium value. The anisotropy of the neutrino flux therefore leaves a robust polar–equatorial imprint on the composition of the ejecta, which is expected to affect both the nucleosynthesis and the kilonova emission.
Among recent studies, Riccigliano et al. (2024) investigated the impact of the nuclear EOS on the nucleosynthesis and kilonova emission from BNS mergers using tracer particles extracted from simulations employing an M0 neutrino transport scheme. They showed that both the isotopic yields and the kilonova emission depend sensitively on the EOS, highlighting significant degeneracies between merger dynamics and nuclear microphysics. More recently, the production of helium (He) and other light elements in neutrino-irradiated winds from mergers with long-lived remnants has gained significant attention because of their potential signatures in kilonova spectra ([78, 108, 109, 5, 13]). In particular, He abundance has been proposed as a diagnostic of the remnant lifetime and, consequently, of the nuclear EOS, since long-lived remnants are expected to synthesize substantially more He than promptly collapsing systems ([109]). More generally, these developments highlight the need for nucleosynthesis calculations based on numerical-relativity simulations with accurate modelling of neutrino irradiation, and that provide detailed isotopic yields for both the dynamical and disc wind ejecta. This need is reinforced by recent long-duration simulations, which demonstrated that late-time disc outflows can significantly modify the final abundance pattern compared with simulations terminated shortly after merger, emphasizing the importance of modelling the long-term evolution of the remnant and its winds (e.g. [30, 31, 53, 27, 111, 82, 60]).
In this paper, we present publicly available -process nucleosynthesis yields from BNS mergers with long-lived remnants, obtained for the first time by post-processing tracer particles extracted from simulations in full general relativity employing the M1 neutrino transport scheme. Our sample spans two nuclear EOS, a range of binary total masses, and different mass ratios enabling us to investigate how the properties of the binary influence the resulting nucleosynthesis. The neutrino emission and the electron fraction evolution of these simulations were analysed in detail in Çokluk et al. [12]. In this work, we focus on the nucleosynthesis. Elemental and isotopic yields for each simulation are obtained by processing tracer particles with the WinNet nuclear network ([92]). Since neutrino and viscosity driven winds are expected to unbind up to ∼40% of the remnant accretion disc mass (e.g. [105, 30, 31, 27]), we provide nucleosynthesis yields for multiple prescriptions of the wind ejecta mass to account for the current uncertainties associated with the late-time disc outflows. The complete set of isotopic and elemental yields, together with the corresponding tracer trajectories, is publicly released through the MARTINI platform ([10]), providing a new resource for kilonova modelling, Galactic chemical evolution studies, and comparisons with the chemical abundance patterns observed in metal-poor and actinide-boost stars.
The paper is structured as follows. In Sect. 2, we describe the numerical relativity simulations, the nucleosynthesis calculations, and the methodology followed to compute the nucleosynthesis yields. In Sect. 3, we present the properties of the merger ejecta, the evolution of the tracer particles, and the resulting yields at 1 day and 1 Gyr after the merger. In Sect. 4, we compare our results to several studies in the literature. In Sect. 5, we describe the implementation of the computed yields within the MARTINI platform and provide an example of data use. Finally, Sect. 6 summarises our main results and presents our conclusions.
Method
BNS merger simulations
We perform eight BNS merger simulations in full general relativity, spanning different component masses and two nuclear EOSs. The key properties of each simulation are summarised in Table 1, where denote the gravitational masses of the two NSs measured at infinity, is the binary mass ratio, is the total mass of the binary, is the merger time, and is the final simulation time. In all cases, the merger results in a long-lived NS remnant (i.e. a NS surviving for at least 50 ms – 80 ms after the merger), surrounded by a matter disc.
| EOS | [] | [] | [] | [ms] | [ms] | Simulation name | |
| DD2 | 1.0899 | 1.5570 | 0.70 | 2.6469 | 9.17 | 70.99 | DD2_M10899-15570_M1 |
| DD2 | 1.1968 | 1.4080 | 0.85 | 2.6048 | 10.08 | 65.33 | DD2_M11968-14080_M1 |
| DD2 | 1.2075 | 1.7250 | 0.70 | 2.9325 | 7.42 | 73.83 | DD2_M12075-17250_M1 |
| DD2 | 1.2980 | 1.2980 | 1.00 | 2.5960 | 10.18 | 73.14 | DD2_M12980-12980_M1 |
| DD2 | 1.3260 | 1.5600 | 0.85 | 2.8860 | 7.83 | 74.65 | DD2_M13260-15600_M1 |
| DD2 | 1.4350 | 1.4350 | 1.00 | 2.8700 | 7.88 | 88.90 | DD2_M14350-14350_M1 |
| SFHo | 1.2980 | 1.2980 | 1.00 | 2.5960 | 11.59 | 86.00 | SFHo_M12980-12980_M1 |
| SFHo | 1.2980 | 1.2980 | 1.00 | 2.5960 | 12.38 | 63.55 | SFHo_M12980-12980_M1_HR |
Table 1. List of the BNS merger simulations considered in this work.
Each simulation is initialised using the Lorene library [35, 34] with two irrotational NSs in quasi-circular orbit placed at a separation of 40 – 50 km, corresponding to a few orbits before merger. The two NSs are in neutrino-less beta and hydrostationary equilibrium. The initial configuration is then evolved by solving Einstein’s equations coupled to general relativistic hydrodynamics in a dimension spacetime. The evolution of spacetime is performed through the CTGamma code [94], which is part of Einstein Toolkit [59, 26], employing the Z4c formulation of Einstein’s equations [7, 38], the 1+log condition for the lapse and the integrated Gamma-driver condition for the shift. Fourth-order finite-difference operators are used to evolve the metric fields, applying the Kreiss–Oliger dissipation to ensure the nonlinear stability of the evolution. The general relativistic hydrodynamics simulation of matter evolution is performed through the WhiskyTHC code [87, 88 b,c] adopting flux-conservative high-resolution shock-capturing schemes. We use fifth-order monotonicity-preserving reconstruction [113] in combination with an HLLE approximate Riemann solver [36, 25]. The hydrodynamic evolution and spacetime dynamics are coupled through the method of lines adopting a third order Runge–Kutta scheme for time integration with a fixed Courant–Friedrichs–Lewy factor of 0.075. The computational domain is discretised using Carpet adaptive mesh refinement [102] (Reisswig et al. 2013a) with seven levels of nested Cartesian grids and a factor-of-two refinement between successive grid levels. The finest refinement level covers both NSs during the inspiral and the central remnant after merger. The numerical resolution is set by the grid spacing of the innermost level, which is m for all models. To assess the impact of the grid spacing on the nucleosynthesis outcome, we also evolve the SFHo model with at a higher resolution of m (model SFHo_M12980–12980_M1_HR in Table 1). The computational domain extends to km in each spatial direction. To guarantee numerical stability in low-density regions, we set an artificial atmosphere throughout the computational domain with floor density g cm.
Nuclear matter is modelled using finite-temperature, composition-dependent EOSs, specifically the HSD(DD2) EOS [115, 37 hereafter DD2] and the SFHo EOS [112]. The modelling of nuclear interactions in both EOSs is based on a relativistic mean field theory, and the relevant degrees of freedom are neutrons, protons, nuclei, electrons, positrons, and photons. Both EOSs are consistent with current experimental constraints on nuclear matter properties [104, 73, 23] (Danielwicz & Lee 2014) and are in agreement with present astrophysical constraints on the NS maximum mass and radii [3, 58, 73, 63, 64, 81] (Riles et al. 2019). SFHo produces more compact NSs compared to DD2: for cold, non-rotating NSs, SFHo yields a maximum mass of and corresponding radius km, both smaller than the DD2 values of and km. These two EOSs allow us to span a relatively broad range of NS compactness without present uncertainties.
Neutrino transport is handled with a truncated, two-moment energy-integrated radiation transport scheme through the module THC_M1 [85] implemented in WhiskyTHC. The scheme evolves the first two energy-integrated moments of the neutrino distribution function (energy density and momentum density flux) retaining all nonlinear neutrino-matter coupling terms. The Minerbo closure is applied, ensuring consistency between neutrino diffusive and free-streaming regimes. We take into account three independent neutrino species: electron neutrinos , electron antineutrinos , and a collective species , representing muonic and tauonic neutrinos and antineutrinos. Weak interaction rates include charged-current, thermal, and scattering processes as described in previous works (e.g. [32, 76]).
Ejected matter
To compute the properties of the ejected matter, we combine two complementary approaches. The first involves computing the flux of ejected matter through a spherical surface located at a radius of km from the center of mass of the system. We define ejected matter as matter that satisfies the Bernoulli criterion , where is the time component of the four velocity and the specific enthalpy, and has a positive radial velocity at the detector surface . We compute the total ejected mass, , by integrating the flux from the merger time, , to the end of the simulation, . To distinguish between dynamical and wind ejecta, we analyse the mass ejection rate, , at . We define the dynamical ejecta mass, , as the mass accumulated up to the first minimum following the first maximum in . The wind ejecta mass is defined as the remainder of the ejected matter. We notice that with this definition, the dynamical ejecta mass computed via the Bernoulli criterion practically coincides with that derived from the geodesic criterion (i.e. ). However, as expected, the subsequent wind ejecta mass is significantly larger when using the Bernoulli criterion compared to the geodesic one (e.g. [29]).
Since in all our models the ejecta mass has not saturated by the end of the simulation, the computed wind mass, and consequently the total ejected mass, should be considered as lower limits. However, since the remnant NS has not collapsed yet to a BH, we extrapolate the properties of the ejecta over timescales longer than the simulated ones, but comparable to the disc lifetime, by assuming that the disc wind has reached a stationary state. The additional wind ejecta mass is estimated as a varying fraction of the disc mass, spanning present uncertainties in the long-term post-merger remnant evolution:
where is the wind ejecta mass computed up to the end of the simulation, is the disc mass at time , is Notes. The equation of state is listed under EOS. and denote the gravitational masses of the binary components, is the mass ratio, is the total mass of the binary, is the merger time, and is the final simulation time.
the portion of the disc mass satisfying the Bernoulli criterion at . In each simulation, is calculated by integrating the conserved density over the computational domain in the region where the rest mass density . Table 2 lists, for each simulation, the dynamically ejected mass and the final disc mass. It also reports the wind ejecta mass for the different wind prescriptions, with minimum and maximum , and the corresponding total ejected mass
| Simulation name | ||||||
| DD2_M10899-15570_M1 | 1.772 | [16.86, 126.7] | 0.307 | [18.63, 128.4] | [0.833, 0.437] | 4530 |
| DD2_M11968-14080_M1 | 1.746 | [7.086, 110.1] | 0.283 | [8.833, 111.8] | [0.632, 0.224] | 3417 |
| DD2_M12075-17250_M1 | 1.160 | [10.82, 116.4] | 0.284 | [11.98, 117.6] | [0.616, 0.267] | 4032 |
| DD2_M12980-12980_M1 | 1.049 | [6.147, 89.38] | 0.232 | [7.197, 90.43] | [0.595, 0.481] | 4086 |
| DD2_M13260-15600_M1 | 1.136 | [10.81, 107.5] | 0.260 | [11.94, 108.6] | [0.531, 0.099] | 3415 |
| DD2_M14350-14350_M1 | 0.599 | [9.609, 77.81] | 0.182 | [10.21, 78.41] | [0.407, 0.168] | 3321 |
| SFHo_M12980-12980_M1 | 1.208 | [8.111, 83.05] | 0.197 | [9.319, 84.26] | [0.456, 0.103] | 2861 |
| SFHo_M12980-12980_M1_HR | 1.121 | [6.026, 74.65] | 0.185 | [7.147, 75.77] | [0.490, 0.118] | 3011 |
Table 2. Summary of the outcome of the BNS merger simulations considered in this work.
with minimum and maximum .
The second approach involves tracking passively advected tracer particles. Each simulation initializes tracers approximately 2 ms prior the merger, distributed uniformly in space within a radius . Each tracer records thermodynamic and kinetic variables every ms, and in particular temperature , entropy , rest mass density , radius , radial velocity , and electron fraction . We define tracers associated with the ejecta as all tracers crossing that satisfy the Bernoulli criterion and have outside .
Tracer particles selection
To perform the nucleosynthesis calculations, we downsample the full tracer dataset by selecting 3000 – 5000 representative tracers for each merger model. This sample size is sufficient to accurately reproduce the nucleosynthesis properties of the outflow, provided that the tracers are properly sampled [95]. To this end, we developed a sampling algorithm designed to preserve the properties of the ejecta.
First, we divide the outflow flux through into eight to ten time windows, . We set ms during the dynamical phase and ms during the wind phase. Within each time window, we record the entropy, , and the electron fraction, , of all ejecta crossing the detector surface. These data are then binned in the plane using 28 bins linearly spaced in and 25 bins logarithmically spaced in . Next, we identify all tracers that cross the detector surface within the given time window and satisfy the Bernoulli criterion. Across all simulations, the number of tracers crossing the detector during the wind phase is , corresponding to an average tracer mass of . Therefore, we retain all wind-phase tracers without subsampling. During the dynamical phase, however, the number of crossing tracers is up to two orders of magnitude larger, so we subsample them as follows. Given the total mass ejected during the dynamical phase, we first estimate the target number of dynamical tracers to retain, , by requiring an average tracer mass of . Each dynamical-phase tracer is then assigned, within its crossing-time window, to a bin based on its values at the crossing time. For each dynamical-phase time window, we define the number of tracers to select as
is the number of bins occupied by at least one tracer in that window, and is the mass ejected within the window. We first select one tracer at random from each occupied bin, guaranteeing full coverage. If , we then randomly select additional tracers, weighting the selection probability by the ratio , where is the bin mass. This procedure preserves the - distribution of the outflow. We verified that this selection also provides a consistent sampling of the ejecta distributions in polar angle, , and radial velocity, . We find that the tracer temperature at the detector surface is always GK. The number of selected tracers, , for each simulation is reported in Table 2.
To compute the nucleosynthesis yields from each BNS merger model, we assign a representative mass to each selected tracer. For each time window, we bin the ejected mass in the three-dimensional parameter space at , where the polar angle is defined such that deg at the pole and deg at the equator. We adopt a coarser binning than in the subsampling step above, to ensure adequate tracer statistics per three-dimensional cell: 14 bins linearly spaced in , 10 bins logarithmically spaced in , and 26 bins linearly spaced in deg. The selected tracers are assigned to the corresponding bins according to their properties at . If a massive cell contains no tracers, we redistribute its mass among the neighboring cells in the three-dimensional parameter space until every massive cell is populated by at least one tracer. The mass associated with each populated cell is then divided equally among the tracers it contains. We validate this mass-assignment procedure by comparing the ejecta mass distributions as functions of , , , and obtained directly from the outflow flux with those reconstructed from the mass-weighted tracers. The results, shown in Fig. A.1, demonstrate a good agreement between the two approaches.
Eq. (1) introduces five estimates of the total wind ejecta mass. The first () uses only the mass directly ejected during the simulation, while the remaining estimates () include an additional contribution from the accretion disc. For , the excess mass, , is distributed among the tracers ejected during the final available time window, corresponding to the last 10 ms of the simulation.
Notes. , , , and represent the masses of the dynamical ejecta, wind ejecta, disc, and total ejecta. For the wind ejecta and the total ejecta, we provide the minimum and maximum amount of ejected matter, as discussed in Sect. 2.2. and are the mass fractions of -process nuclei () in the total ejecta at Gyr, for the minimum () and maximum () ejected mass, respectively. Finally, represents the number of selected tracers.
Nucleosynthesis calculations
To analyse nucleosynthesis, we post-process the thermodynamic evolution of the selected tracers through the open-source nuclear reaction network WinNet (Reichert et al. [92]) and evolve abundances up to Gyr. We provide WinNet with the time-dependent evolution of temperature, density, radius, and electron fraction for each tracer. We set GK as the threshold temperature below which the full network nucleosynthesis calculation starts, with the initial composition determined by nuclear statistical equilibrium (NSE). From that point on, the network follows the tracer’s hydrodynamical density and radius until the end of the trajectory. Afterwards, the density decreases as and the radius grows linearly at the velocity of the last hydrodynamical output interval, i.e. we assume homologous expansion until Gyr. The temperature and the composition are evolved by the network. We do not explicitly include neutrino reactions in the network simulation. We notice that the impact of neutrinos in determining the ejecta electron fraction is already accounted for through M1 neutrino transport in the hydrodynamical simulation. When the network calculations are initialised ( GK), the ejecta have expanded and the neutrino flux is significantly reduced, so that neutron absorption reactions have a subdominant effect on the final nucleosynthesis1.
The network includes 8904 nuclei up to . We employ the REACLIB database (Cyburt et al. [18]) as our baseline for reaction rates, prioritizing experimental data whenever available. Nuclear properties are adopted from the AME2020 evaluation (Wang et al. [118]), complemented with theoretical values derived from the FRDM2012 model (Möller et al. [67]). Using the TALYS code (Koning et al. [55]) with recommended parameters from Rochman et al. [97], we consistently calculate neutron-capture rates, as well as , , and channels, based on the same FRDM2012 mass model. Inverse rates are obtained from detailed balance. Beta decay rates are taken from Möller et al. [65]. Additionally, for temperatures larger than GK, we use temperature- and density-dependent theoretical weak interaction rates (including and decays, as well as positron and electron captures) as described in Reichert et al. [92]. Furthermore, -delayed fission rates and neutron-induced fission rates are taken from Panov et al. [74] and Panov et al. [75], respectively. Spontaneous fission rates are obtained using the semi-empirical formula of Khubayvaatar (2020), adopting the fission barriers provided in Möller et al. [66]. Fission fragment distributions are adopted from Mumpower et al. [68] for -delayed and neutron-induced fission, and from Kodama & Takahashi [54] for spontaneous fission.
In Table 2, we report the mass fraction of -process nuclei () in the total ejecta at Gyr, for the minimum () and maximum () wind ejecta mass, denoted and , respectively.
Results
Ejected matter
The time evolution of the ejecta properties reveals several common features across all simulations, particularly in the transition from dynamical to wind ejecta. This evolution is summarised in Figure 1 and Figure 2 for symmetric () and asymmetric () BNS models, respectively. Both figures show the median values and the 16th to 84th percentile distributions of the electron fraction , entropy , radial velocity , and polar angle as functions of the post-merger time.

Figure 1. Evolution of the properties of the ejected matter for the simulations as a function of post-merger time , recorded at the detector surface km. Each marker represents the median value of the electron fraction (top left), specific entropy (top right), radial velocity (bottom left), and polar angle (bottom right) calculated within individual time windows . Markers are positioned at the centre of their respective time windows. Marker colours indicate the BNS merger simulation while marker shapes indicate the amount of mass ejected during each time interval. Vertical error bars indicate the 16th and 84th percentiles of the distribution within each window. Colours refer to different simulations while markers denote the range of mass ejected during each interval.

Figure 2. Same as in Figure 1, but for BNS simulations with asymmetric mass ratio ().
In all models, the dynamical ejecta are initially ( ms) characterised by a small amount of mass (), expanding at high speed () across the equatorial plane, with a high median electron fraction around , and a relatively high entropy ( baryon and baryon for and , respectively). These ejecta originate from the collision of the two NSs.
After that, but still within the first ms after merger, the median electron fraction and the expansion velocity decrease, accounting for the ejection of slower and more neutron rich matter due to tidal interaction and remnant shocks. The minima of are consistently reached during this phase. Specifically, we find that all simulations reach , with the exception of DD2_M14350-14350_M1 and SFHo_M12980-12980_M1_HR. The lowest values () are observed in the two highly asymmetric cases (), namely DD2_M10899-15570_M1 and DD2_M12075-17250_M1, followed by the model DD2_M13260-15600_M1. This behaviour is consistent with previous analysis, showing that the competition between the expansion timescale and the neutrino irradiation effect eventually sets the ejecta composition. Since dynamical ejecta are mostly emit-
ted across the equator (in particular, for very asymmetric models), where the shielding effect due to the presence of the disc is stronger, most of the ejecta do not reach the equilibrium electron fraction, partially preserving their initial neutron richness. In the subsequent wind phase, steadily increases again, finally stabilising at around 30 ms after merger. At the same time, both the median entropy and radial velocity stabilise around and , respectively, during wind ejection. This entire trend shows the growing impact of the neutrino irradiation, which progressively increases the electron fraction of the outflowing matter. In particular, neutrino absorption provides enough energy to fluid elements to expand and decrease temperature inside the outer disc layers. The subsequent nuclear recombination supplies the necessary energy to unbind matter, setting the final speed and entropy. Regarding the angular distribution, most of the ejecta progressively migrate toward more polar directions once the neutrino-driven wind develops (cf. Bernuzzi et al. [8]), ultimately stabilising at with a relatively broad dispersion ().
When comparing SFHo_M12980-12980_M1 and DD2_M12980-12980_M1, in the wind ejecta SFHo yields higher electron fraction values and a significantly smaller spread compared to DD2. Notably, while the baseline and high-resolution (HR) runs of the SFHo model show distinct electron fraction behaviours during the dynamical phase, they converge to nearly identical values during the wind phase. Overall, the dependence of the electron fraction on the binary mass ratio and EOS, together with its systematic increase during the wind phase due to neutrino irradiation, is consistent with previous BNS merger simulations employing detailed neutrino transport ([82, 83, 8]). Also the variation due to numerical resolution is consistent with previous studies (see e.g. Zappa et al. [122]).
In Figure 3, we show the mass distribution of , , , and for the dynamical (solid lines) and wind (dashed lines) ejecta. The largest dynamical ejecta mass is found in the asymmetric (), relatively low-mass model DD2_M10899-15570_M1, while the smallest is produced by the equal and high-mass model DD2_M14350-14350_M1. For most simulations, the electron fraction of the dynamical ejecta peaks at . However, both the SFHo models peak at higher values around . Since SFHo is a softer equation of state, it leads to stronger shocks and higher remnant temperatures, enhancing positron captures on neutrons and increasing neutrino fluxes, therefore shifting the distribution toward higher values. The equal-mass model employing the DD2 EOS with exhibits a bimodal distribution with peaks at 0.1 and 0.4. The first peak originates from the tidal interaction between the two merging stars, enhanced by the EOS stiffness. The second one, coming from the NS contact interface, is amplified by the larger NS masses and by the symmetric merger. This bimodality is mirrored in its entropy distribution, which shows peaks at and , while all other models peak near . The velocity distribution of dynamical ejecta typically peaks at , though SFHo_M12980-12980_M1 and DD2_M14350-14350_M1 exhibit significant tails reaching up to . While most matter is emitted near the equatorial plane, the high resolution SFHo displays a broader emission angle, extending toward the poles down to .

Figure 3. Dynamical (solid lines) and wind (dashed lines) mass distribution in , , , and for all simulations reported in Table 2.
During the wind phase, the electron fraction distribution generally peaks between for all simulations. The two asymmetric DD2 models () are notable exceptions, showing a double-peaked distribution ( and 0.5). Concerning the entropy in the wind, all simulations show a peak near . The velocity distribution peaks at , with the wind phase generally being slower than the dynamical one. Finally, the wind shows peaks at for most models, driven by neutrino irradiated winds that shift the emission toward the poles. However, the models retain significant equatorial emission, due to their tidally dominated merger dynamics. Specifically, the DD2 simulation with and peaks exclusively at the equator, while DD2 with and , shows a double peak at the equator and around .
Tracers evolution
In Figure 4, we show the time evolution of the electron fraction (left), radial velocity (center), and specific entropy (right) for the tracer particles extracted from the simulation SFHo_M12980-12980_M1_HR. In each panel, we include the trajectories that crossed the detector surface, , within the time window indicated by the light grey region. Thus, the first three rows represent tracers ejected during the dynamical phase, while the last three rows refer to the early wind phase, up to ms after merger. For each time window, three representative tracer trajectories are superimposed in green, blue, and magenta. Tracer trajectories crossing between ms and ms after merger (end of the simulation) behave as those represented in the last row.

Figure 4. Ejected tracer trajectories (grey) for the electron fraction (left), radial velocity (center), and specific entropy (right) as functions of , for the simulation SFHo_M12980–12980_M1_HR. Each row corresponds to a time window , marked by the shaded grey region bounded by solid black lines; within each row, we plot the trajectories of all tracers crossing the detector surface within that window. The blue, magenta, and green lines highlight three representative tracers. The first three rows represent tracers crossing during the dynamical phase, while the remaining rows refer to the wind phase up to ms after merger.
Tracers ejected immediately after merger (first row) originate from the shock interface and exhibit the characteristic behaviour of shocked matter: the radial velocity shows jumps due to shocks and peaks at , while the entropy rises above . The remaining dynamical ejecta (second and third rows) are characterised by tracers spanning a broad range of electron fractions, . These tracers show a rapid velocity increase immediately after merger, followed by an approximately constant radial velocity, and a more gradual rise in entropy. The representative trajectories in these rows highlight different behaviours. Some tracers are ejected ballistically (e.g., blue in the second row), maintaining nearly constant , , and after the initial acceleration. Other tracers show signatures of fluid compression (e.g., the magenta and blue tracers in the third row), leading to increases in both and . A further class of tracers, characterised by multiple oscillations in , is ejected after spending time in the forming remnant disc (e.g., the green tracer in the third row). These tracers exhibit increasing and due to neutrino irradiation. The impact of neutrino irradiation becomes increasingly significant with the onset of the wind phase, leading to a systematic rise in both and as tracers spend time in the disc. Radial velocity oscillations around zero reflect the back-and-forth motion of tracers within the disc prior to their ejection (see, e.g., the green and magenta tracers in the fifth and sixth rows). This behaviour becomes progressively more common at later times.
Tracer trajectories in all other simulations follow similar trends; however, during the dynamical phase, simulations with highly asymmetric mass ratios exhibit a distribution that is more strongly weighted toward lower values (see also the discussion in Sect. 3.1).
In Figure 5, we present the tracer distribution in the plane, extracted at a time , defined such that the temperature 1 GK for all , considering the simulation SFHo_M12980-12980_M1_HR. The upper (lower) row corresponds to tracers that crossed the detector surface during the

Figure 5. Tracers as function of the radial velocity, , and the polar angle, , extracted at time , such that the temperature GK for all , considering the simulation SFHo_M12980-12980_M1_HR. Upper (lower) row represents tracers crossing the detector surface during the dynamical (wind) phase. Colour bars refer to the electron fraction (left) and the entropy (right).
dynamical (wind) phase. The colour of each tracer represents its electron fraction (left panels) and entropy (right panels). A clear trend emerges in the tracer distribution: tracers with higher electron fraction and entropy are preferentially emitted near the polar regions and exhibit larger radial velocities (e.g. Radice & Bernuzzi [83]; Bernuzzi et al. [8]). As the electron fraction and entropy decrease, the tracers shift toward the equatorial regions and their radial velocities progressively decrease. This trend is particularly evident for tracers emitted during the wind phase, whereas tracers associated with the dynamical ejecta show a much broader spread2. We also note the presence of dynamical ejecta tracers in the high-velocity tail, reaching , consistently with the findings of Radice et al. [86]; Nedora et al. [70]; Bernuzzi et al. [8]; Rosswog et al. [100]. Tracers from the other simulations display the same qualitative correlations between , , , and , although the detailed shape of the distributions varies from simulation to simulation3.
Nucleosynthesis
Yields at 1 day
The analysis of AT2017gfo spectra revealed a robust identification of strontium (Sr) features (Watson et al. [119]; Domoto et al. [21]), with an ongoing discussion on the potential contribution of He (Perego et al. [78]; Sneppen et al. [108, 109]; Chiba et al. [13]), and plausible identifications of features due to yttrium (Y) (Sneppen & Watson [110]), tellurium (Te) (Hotokezaka et al. [45]; Gillanders et al. [33]), lanthanum (La) and cerium (Ce) (Domoto et al. [20]), and gadolinium (Gd) (Gillanders et al. [33]; Rahmouni et al. [91]). In addition, the effect of nickel (Ni) (Jacobi et al. [46]) and Th (Domoto et al. [22]) on kilonova lightcurves and spectra was also recently investigated.
In Figure 6, we show the masses of He, Ni, Sr, Y, Te, La, Ce, Gd, and Th present at day after merger in the ejecta of all simulations in our sample. The figure includes the three of the five prescriptions for the wind ejecta mass introduced in Eq. (1). We first consider the case , ( , leftmost panel), where only the ejecta directly produced during the simulation are included. In this case, all merger models predict similar yields of the lightest elements, namely He (–), Ni (–), Sr (–), and Y (–). In contrast, the production of heavier elements, including Te, La, Ce, Gd, and Th, exhibits a much larger model-to-model variation, spanning more than one order of magnitude. As expected, unequal-mass BNSs generally produce larger amounts of heavy elements than equal-mass systems. As discussed in Sect. 3.1 and shown in Figure 3, these mergers eject a larger fraction of low and low entropy material due to the reduced contribution of shock-heated ejecta. Such thermodynamic conditions favour the production of heavy -process nuclei. Among the elements shown in Figure 6, He and Ni are the most abundant in all models, with the exception of DD2_M10899-15570_M1, which synthesises more Te () than either He or Ni.

Figure 6. Masses of selected elements synthesised at day after merger for all simulations listed in Table 2. From left to right, the wind ejecta mass is computed according to Eq. (1), by considering only the wind mass directly ejected during the simulation time () and by adding 20% or 40% of the accretion disc mass ( and ), respectively.
Increasing the wind mass () enhances the production of all elements. The effect is particularly pronounced for He, Ni, Sr, Y, and Te, while the spread among different merger models also increases. The origin of this behaviour becomes evident from Figure 7, where we show the ratio of the elemental yield in the dynamical ejecta to that in the wind ejecta in the case . For every simulation, most of the He, Ni, Sr, Y, and Te are synthesised in the wind ejecta. Consequently, increasing the wind mass primarily boosts the production of these elements. Interestingly, He exhibits only a small model-to-model variation in both the dynamical and wind components, indicating that its production is relatively insensitive to the details of the merger. Focusing on nuclei heavier than Te, for most simulations they are predominantly synthesised in the dynamical ejecta and in the early-time wind ejecta, where the material is characterised by lower entropy and lower electron fraction. As a result, increasing the late-time wind mass has only a modest impact on their total yields. However, the two unequal-mass mergers with , namely DD2_M10899-15570_M1 and DD2_M12075-17250_M1, produce a larger amount of heavy elements in the wind than in the dynamical ejecta, with a significant component of heavy elements coming from the late time wind. The equal-mass model DD2_M12980-12980_M1 behaves similarly, synthesizing comparable amounts of heavy nuclei in the two ejecta components, again with a significant late-time wind contribution. This behaviour is consistent with Figure 1 and Figure 2, which show that the late-time wind ejecta of these three models retain a broad distribution extending to low values. Consequently, increasing the wind mass also significantly enhances the production of heavy nuclei in these simulations, amplifying the model-to-model spread seen in Figure 6. This broad distribution in the wind ejecta of the more asymmetric simulations is consistent with the spiral-wave wind mechanism, which produces a broad distribution and is particularly efficient in more asymmetric mergers (Nedora et al. [70]). The less strongly asymmetric simulations DD2_M11968-14080_M1 and DD2_M13260-15600_M1 follow a different pattern: although the amount of heavy nuclei produced in the dynamical ejecta is comparable to or smaller than that produced in the wind ejecta, most of the heavy-nuclei production in the wind occurs at early times. Increasing the late-time wind component therefore does not significantly increase the yield of heavy nuclei in these two models.

Figure 7. Ratio of the mass of selected elements synthesised in the dynamical ejecta, , to that synthesised in the wind ejecta, , at day after merger, for all simulations listed in Table 2. The wind ejecta mass is computed according to Eq. (1) considering only the material directly ejected during the simulation (). The legend is identical to that of Figure 6.
The thermodynamic conditions under which the different elements are synthesised are illustrated in Figure 8. The plots show the tracer distribution in the - planes at time for the representative simulation SFHo_M12980-12980_M1_HR. They provide a direct visualization of the distinct thermodynamic regimes responsible for the synthesis of light and heavy nuclei. He is predominantly synthesised in high- () and high-entropy () ejecta, as expected for -rich NSE freezeout. Ni is also mainly produced in the high- region (), but a significant fraction also originates from the ejecta with slightly lower electron fractions (–) and lower entropies (–). The production of Sr and Y extends over a much broader region of parameter space, approximately spanning – and –. In this region, the dominant contribution follows a correlation between the electron fraction and the entropy, with higher electron fractions corresponding to higher entropies. In contrast, the production of heavier nuclei is concentrated in low-, low-entropy ejecta, with and . This behaviour is particularly evident for Th, whose production peaks in the region and , in agreement with the results of Eichler et al. [24].

Figure 8. Tracer distribution in the - plane at time , defined such that GK for all , for the simulation SFHo_M12980-12980_M1_HR. The color scale indicates the mass of the element produced by each tracer at day after merger. The element corresponding to each panel is indicated in its title.
The He and Sr mass fractions in BNS merger ejecta have recently attracted considerable attention, owing to their impact on early kilonova spectra and their sensitivity to the lifetime of the merger remnant and the strength of neutrino-driven winds (e.g. [78, 109, 13, 5]). In particular, the abundances of He and Sr depend strongly on the thermodynamic properties of the ejecta, especially the electron fraction, making them valuable probes of the post-merger evolution. To facilitate future studies of the connection between ejecta coming from long-lived merger remnants and kilonova observables, we report the He and Sr mass fractions, and respectively, predicted by our simulations. In Figure 9, we show as a function of the radial velocity and polar angle at day after merger. As described in Sect. 2.4, after the end of the hydrodynamical trajectory the network assumes homologous expansion at the velocity of the last hydrodynamical point. We therefore take of each tracer to be its velocity at the last hydrodynamical point, and to be its polar angle at the same point, assuming that both remain constant during the subsequent expansion. We then construct a two-dimensional map of by binning all tracers on a grid of 10 velocity bins, linearly spaced in , and 11 polar-angle bins, equally spaced in between the pole and the equator. In each cell, we compute the as

Figure 9. He mass fraction, , computed at day as a function of expansion velocity, , and polar angle, , for all simulations listed in Table 2. In each bin bin, is computed as in Eq. (4).
where the sums run over all tracers in that cell, is the mass of tracer , and is the single tracer He mass fraction. We find that follows a similar pattern in all our simulations: is largest, reaching –, in the polar region at , and decreases gradually toward lower velocities and the equatorial region, dropping below for near the equator. Notably, across the typical photospheric velocity range in all simulations, consistent with the predictions of Sneppen et al. [109] for mergers producing long-lived remnants. The one exception is the region very close to the equator (), where the He fraction falls below 0.1 for some simulations.
In Figure 10, we compare the He and Sr mass fractions, and , respectively, for the dynamical and wind ejecta in the case. The mass fractions are defined as the mass of He or Sr synthesised in each ejecta component divided by the mass of the corresponding ejecta component. In all simulations, the He mass fraction exceeds that of Sr, , in both ejecta components. The dynamical ejecta exhibit a relatively large model-to-model scatter. Therefore, for a fixed EOS, the binary mass ratio and total mass have significant impact on the He and Sr mass fractions in the dynamical ejecta, with possible implications on kilonova spectral features sensitive to these quantities. On the contrary, the wind ejecta cluster around – and , with the exception of DD2_M10899-15570_M1, indicating a moderate dependence of these quantities on the BNS system total mass and mass ratio. The total He mass fraction (dynamical plus wind) is at least 0.1 in all simulations, with the minimum, , reached by DD2_M10899-15570_M1.

Figure 10. Sr mass fraction as a function of the He mass fraction for the dynamical ejecta (upper panel) and the wind ejecta (lower panel) in all simulations listed in Table 2. The legend is the same as in Figure 6.
Yields at 1 Gyr
In Figure 11, we show the nuclear yields at Gyr for all simulations listed in Table 1, grouped into Hydrogen (H), He, Fe-peak, first-peak, second-peak, rare-earth, third-peak, and trans-lead nuclei. Open bars denote the total mass synthesised in each group in the case , while filled bars indicate the contribution of the dynamical ejecta alone. All models produce similarly large total masses () of He, Fe-peak, and first-peak nuclei, whereas the masses of second-peak, rare-earth, third-peak, and trans-lead nuclei vary substantially among models, particularly in the wind ejecta. The yields of H also depend strongly on the BNS model, in both the dynamical and wind ejecta.

Figure 11. Nuclear yields computed at Gyr for all simulations listed in Table 1. Open bars denote the total mass synthesised in each elemental group, while filled bars indicate the component from the dynamical ejecta. The Fe peak includes elements with . The first-peak, second-peak, rare-earth, third-peak, and trans-lead groups comprise isotopes with mass numbers , , , , and , respectively.
The final abundance patterns are compared with the solar -process residuals (Prantzos et al. [80]) in Figure 12, normalising the solar residuals to match the mass of the isotope in each simulation. We show in green and light blue the contributions from the dynamical and wind ejecta, respectively, and in blue the sum of the two. All simulations successfully reproduce the second and third -process peaks, while exhibiting more pronounced differences in the first peak, the rare-earth (lanthanide) region, and the mass range between the first and second peaks. These variations primarily reflect the different amounts of low- material ejected by the various merger models and the relative contributions of the dynamical and wind ejecta to the overall nucleosynthesis. For symmetric BNS systems, the dynamical ejecta dominates the production of isotopes above the second peak (), while the wind ejecta dominates the production of isotopes below the second peak. The exception is DD2_M12980-12980_M1, for which the dynamical and wind ejecta contribute comparably above the second peak. For asymmetric mergers, instead, the largest contribution to the nucleosynthesis of all isotopes comes from the wind ejecta, since these systems have larger disks with a significant wind-ejecta component at . An exception is DD2_M11968-10840_M1, which instead behaves like the symmetric simulations in terms of the division between dynamical and wind ejecta. This can be explained by the smaller wind ejecta of this simulation, whose wind phase lasts 5–10 ms less than in the other asymmetric simulations.

Figure 12. Final isotopic mass distribution for all simulations listed in Table 2, considering the contribution from dynamical ejecta (green), wind ejecta (light blue), and the sum of the two (blue), compared to the solar -process abundance (black dots) from Prantzos et al. [80], normalized to .
Considering the total ejecta (blue curve), we see that while all our simulations successfully reproduce the second and third solar -process peaks, they fail to reproduce the first peak. An exception is DD2_M14350-14350_M1, which reproduces the solar abundance reasonably well for , but overproduces the and regions. Among our models, the equal-mass simulations DD2_M12980-12980_M1 and SFHo_M12980-12980_M1_HR provide the closest overall match to the solar -process abundance pattern. The unequal-mass model DD2_M10899-15570_M1 also reproduces the solar distribution remarkably well for nuclei beyond the second peak, although it significantly underproduces nuclei with . The remaining simulations generally fail to reproduce the solar abundances in the lanthanide region, overproducing this mass range relative to the observed pattern. All simulations tend to overproduce , , and relative to the solar -process abundance pattern. However, we note that the abundances of isotopes from the second peak to the trans-lead region depend significantly on several uncertain nuclear inputs used in the -process calculations (especially around and ), and are therefore themselves uncertain [24, 62].
The differences among the nucleosynthesis yields can be quantified by comparing the masses of selected elements synthesised at Gyr. In Figure 13, we show the total mass of Sr, Ba, Eu, and Th for all simulations in our sample, considering three prescriptions for the wind mass as in Eq. (1), that is , 2, 4. While all models predict similar Sr yields (), the synthesised masses of Ba, Eu, and Th vary by more than one order of magnitude across the simulation sample. Interestingly, in the first panel (), the equal-mass models DD2_M12980-12980_M1 and SFHo_M12980-12980_M1 produce very similar yields despite adopting different EOSs. By contrast, changing the masses of the two NSs while keeping the same EOS results in a much larger variation in the predicted abundances. Furthermore, the numerical resolution also has a non-negligible impact: the difference between the SFHo_M12980-12980_M1_HR and SFHo_M12980-12980_M1 simulations is even larger than the effect of changing the EOS. This suggests that numerical resolution can influence the predicted nucleosynthesis yields as strongly as the underlying microphysical model. As expected, increasing the wind contribution (), has a stronger impact on Sr production, increasing its production up to one order of magnitude, but a much more limited effect the production of Ba, Eu and Th, with the exception of simulations with asymmetric mass ratio, coherently with the discussion in Sect. 3.3.1.

Figure 13. Masses of selected elements synthesised at Gyr after merger for all simulations listed in Table 2. From left to right, the wind ejecta mass is computed according to Eq. (1), by considering only the wind mass directly ejected during the simulation time () and by adding 20% or 40% of the accretion disc mass ( and ), respectively.
Discussion
In this section, we compare our results to studies in which a different approach was used for the BNS merger simulation, the neutrino treatment, or the nucleosynthesis calculations.
Kullmann et al. [56] studied -process nucleosynthesis in four BNS merger models using two EOSs, DD2 and SFHo, each with a symmetric (1.35–1.35 ) and an asymmetric (1.25–1.45 ) configuration, all forming a long-lived NS remnant surrounded by an accretion disc. Their simulations used a relativistic smoothed particle hydrodynamics (SPH) code [72, 6], with neutrino transport handled by the Improved Leakage-Equilibration-Absorption Scheme (ILEAS) [4], which refines standard leakage descriptions with a diffusion-time-scale prescription, a neutrino-equilibration treatment for the trapped regime, and a ray-tracing algorithm for neutrino reabsorption near the neutrinosphere. They identified dynamical ejecta as all mass elements located beyond 100 km at 7 ms after merger whose kinetic plus thermal energy exceeds their gravitational binding energy. We notice that the nuclear inputs used in their nucleosynthesis calculations differ substantially from those adopted here, and refer the reader to Kullmann et al. [56] for details. A comparison of the properties of the dynamical ejecta and the associated nucleosynthesis between our simulations and those of Kullmann et al. [56] shows good overall agreement. Specifically, the mass of dynamically ejected matter in our simulations ranges over , lower than but still comparable to their range of . The mass-weighted average electron fraction used as input for the nucleosynthesis calculations (i.e. evaluated once is per-
manently below 7 GK) ranges over in our simulations, in good agreement with the range reported by Kullmann et al. [56]. Our lower values of arise because our sample includes very asymmetric systems, which tend to produce more neutron-rich dynamical ejecta. Even though we adopt different nuclear inputs, the mass fractions of lanthanides plus actinides (), of -process nuclei with (), and of third-peak -process nuclei with () in the dynamical ejecta are all compatible with Kullmann et al. [56]: we find , compared to their ; , compared to their ; and , compared to their . Our results show a larger spread than Kullmann et al. [56], reflecting the broader diversity of masses and mass ratios explored in our simulations: the largest values are found in our most asymmetric binaries, while the smallest come from the equal-mass SFHo (HR) and the equal-mass DD2 model with per star.
In a follow-up work, Kullmann et al. [57] explored the impact of nuclear uncertainties on the nucleosynthesis of BNS merger ejecta by varying the nuclear mass model, -decay rates, and fission properties. In particular, they computed the Th/U ratio for three simulations performed with the SFHo EOS and binary masses of and , modelling the post-merger remnant as a central black hole of surrounded by a disc of , and varying 15 possible nuclear input configurations. They computed from their simulation output, considering only the and isotopes. They found that, while variations in the nuclear input produce a significant spread in the Th/U ratio, changing the binary system has comparatively little effect on the result. Specifically, they found for the system and for the system at Gyr. We find a fully compatible, but much narrower, range of , since we vary only the EOS and mass ratio, and not the nuclear inputs.
Fujibayashi et al. [31] studied the nucleosynthesis of post-merger ejecta from low-mass NS binaries, including two equal-mass BNS systems with the DD2 EOS: one with , and the other with . Their fully general-relativistic simulations included neutrino radiation and viscous hydrodynamics, using an M1 closure scheme for free-streaming neutrinos and a leakage scheme for trapped neutrinos. Both binaries were evolved for s after merger, with Lagrangian tracers per simulation used for nucleosynthesis calculations. Looking at the nucleosynthesis results from these two models, computed using an -viscosity parameter of 0.04, they found a solar -process-like abundance pattern over a wide range of (–200) only when the post-merger elapsed time was below 0.2 s. The subsequent mass ejection added a substantial amount of matter with , so that the abundances of heavier nuclei were lower than the solar -process residuals. The resulting overall patterns were instead consistent with that of the -process deficient metal-poor star HD 122563 (Cowan et al. [14]; Honda et al. [43]; Roederer et al. [99]). Comparing with our simulation DD2_M12980–12980_M1, whose total mass lies between those of the two Fujibayashi et al. [31] models, we find that, for , we reproduce the solar -process abundance well except for the first peak, as shown in Figure 12. Adopting instead prescription – consistent with a central remnant surviving beyond the end of our simulation – we reproduce the solar -process pattern well up to the second peak, but significantly underproduce isotopes with .
Just et al. [50] investigated the nucleosynthesis in BNS merger models with remnant lifetimes of s before BH formation, following the dynamical ejecta, the ejecta launched during the NS-remnant phase (NS-torus ejecta), and the ejecta from the viscous disintegration of the BH torus (BH-torus ejecta). The evolution was followed up to 10 ms after merger with a 3D general-relativistic SPH code and the ILEAS neutrino scheme (Ardevol-Pulpillo et al. [4]). The subsequent post-merger evolution was axisymmetric, performed with the special-relativistic code ALCAR-AENUS (Just et al. [49]) with energy-dependent M1 neutrino transport, and the ejecta were followed up to 100 s. The post-merger evolution extended the classical -viscosity scheme to capture MRI-driven viscosity in both the accretion torus and the NS remnant, with a dedicated parameter, , controlling the remnant viscosity independently of the disc. Tracers were sampled from the final homologous ejecta, with 4000–5000 effective tracers per model for the dynamical component, and their histories were reconstructed by backward integration in time. Nucleosynthesis was computed with two independent nuclear networks. Their sample comprised ten models, with symmetric () and asymmetric () binaries of total mass , all with the SFHo EOS, for which they varied and the -viscosity parameter . Just et al. [50] summarised in their Table 1 the properties of the dynamical, NS-torus and BH-torus ejecta. Since the NS-torus component included the neutrino-driven wind, we compare their dynamical and NS-torus ejecta with our dynamical and wind ejecta, respectively. Because the two studies define the ejecta components differently, this comparison is only approximate. Their dynamical ejecta have masses of –, larger than ours (–), a difference that is probably due, at least in part, to the different definitions of dynamical ejecta. For , our wind ejecta (–) are also systematically smaller than their NS-torus ejecta (–). This is expected, since our simulations end before the collapse of the remnant to a BH in their models. The dynamical ejecta of our simulations span a broader range of electron fractions. Their average (measured at 5 GK) lies in –, while we find – (also evaluated at 5 GK, for consistency with their definition). This reflects the larger variety of total binary masses, mass ratios and EOSs in our sample, against a single EOS and a fixed total mass in theirs. The lanthanide-plus-actinide mass fraction follows the same trend: they find –, whereas we find –. Our lowest value comes from SFHo_M12980-12980_M1_HR, whose dynamical ejecta have . For the NS-torus ejecta, the average electron fractions are similar (– for them, – for us). The lanthanide-plus-actinide fraction, however, is much smaller in their models () than in ours (– for ). Two effects probably contribute: our wind ejecta may include material that they classify as dynamical, and in most of their models the NS survives for more than 100 ms, which increases the contribution of the neutrino-driven wind and lowers . In agreement with this picture, for , which gives more weight to the neutrino-driven wind, our falls to –, closer to their range. A comparison of the final nucleosynthesis yields is not straightforward, because our simulations have no BH-torus ejecta. However, it is worth to note that our simulations predict similar He mass fraction.
Ricigliano et al. [95] investigated the impact of the nuclear EOS on nucleosynthesis by simulating symmetric BNS mergers with NS masses of , using eight different finite-temperature nuclear EOSs. The latter were obtained by systematically varying nuclear properties, such as the incompressibility and the nucleon effective mass at saturation density, in a Skyrme-type interaction model. Our approach differs from theirs mainly in two respects: the tracer method and the neutrino transport scheme. We evolve a large number () of passive tracers, initialised at the beginning of the simulation, select those that are ejected, and subsample 3000–5000 representative tracers per simulation, as described in Sects. 2.2–2.3. Ricigliano et al. [95], instead, placed tracers in the ejecta at the end of each 3D simulation and integrated their trajectories backward in time. Moreover, they adopted an M0 neutrino transport scheme, while we use the more refined M1 scheme. Despite these differences, we find comparable dynamical and wind ejecta masses; the variability in our sample stems mainly from differences in mass ratio, whereas in theirs it stems from the EOS, suggesting that varying the mass ratio and varying the EOS can produce similar levels of variability in the ejecta mass. The properties of the ejected matter are also broadly similar, though with some notable differences. In both our dynamical and wind ejecta, we find a significant component with that is absent in Ricigliano et al. [95]; in the wind ejecta, we also find more mass in the polar region than they reported. Both differences are consistent with the M0 versus M1 neutrino treatment. The resulting nucleosynthesis is likewise similar overall, with a few exceptions. We find that our total He yield computed at one day is significantly larger than that reported by Ricigliano et al. [95], even for the case. Specifically, the He mass synthesised in the dynamical ejecta exceeds their values by a factor of approximately 2–5, while the wind ejecta produce an order of magnitude more He. Comparing our Figure 11 with figure 7 of Ricigliano et al. [95], we find that, while the H and He mass fraction from the dynamical ejecta is comparable between the two studies, the H and He mass fraction from the wind/disc ejecta is an order of magnitude larger in our case, for all our simulations. This difference is consistent with the different neutrino transport schemes adopted in the two studies. The stronger neutrino irradiation in the M1 treatment leads to more proton-rich ejecta, thereby favouring the synthesis of light nuclei such as H, He and Fe-peak nuclei. By comparing the final nucleosynthesis at 1 Gyr, we find that even though we get comparable Sr yields, our simulations systematically produce larger amounts of Eu and Th than Ricigliano et al. [95], in some cases by up to an order of magnitude. This enhancement is particularly pronounced for the unequal-mass systems. Two effects likely contribute to this difference. First, the asymmetric mergers eject a larger fraction of neutron-rich material, favouring the synthesis of nuclei beyond the second -process peak. Second, differences in the adopted nuclear physics inputs, particularly in the calculation of the nuclear reaction rates and fission fragments, can significantly affect the production of lanthanides and actinides
([24, 62]). A systematic investigation of the impact of the nuclear physics inputs on our yields is left to future work. The impact may be explained by different ejecta conditions within a single -process site, such as BNS mergers, or whether they instead require the existence of multiple distinct astrophysical -process sites ([24, 42]).
Nuclear yields in the MARTINI platform
MARTINI4 is an online platform dedicated to stellar nucleosynthesis and the study of kilonova light curves and spectra ([10]). The platform already hosts nucleosynthesis yields from the slow neutron capture process (-process), previously released through the FRUITY database ([16, 17]), together with atomic data for Se and Zr ([10, 9]).
The -process yields from BNS mergers presented in this article are now also available through MARTINI. The platform allows users to visualize the yields online and to download them in .txt or .csv formats. The query interface, shown in Figure 14, enables users to retrieve elemental or isotopic yields for different combinations of binary masses, EOS, numerical resolution, neutrino transport scheme, and wind ejecta mass prescription (cases as in Eq. (1)). The published yields are provided at day after merger, relevant for kilonova studies, and at Gyr, relevant for studies of galactic chemical evolution. Users interested in the set of tracers used to compute the nucleosynthesis and the final yields can download them from https://doi.org/10.5281/zenodo.23689687.

Figure 14. Query form for NSM r-process yields available in the MARTINI platform.
In the following section, we discuss an example of usage of the yields available through the MARTINI platform.
Example of application: chronometer pairs
The nucleosynthesis yields released through the MARTINI platform can be readily evolved beyond Gyr to compute radioactive chronometer pairs, such as Th/Eu, for merger models spanning a wide range of ejecta properties and -process outcomes. These predictions can be directly compared with abundance measurements in metal-poor and actinide-boost stars, providing a valuable tool for investigating the origin of the observed actinide enhancement. In particular, they contribute to the long-standing debate over whether actinide-boost stars can
To compute the chronometer pairs Th/Eu, Th/Dy, and U/Th at Gyr, it is sufficient to evolve the abundances released at Gyr over an additional 9 Gyr, taking into account the radioactive decay of and . The abundance ratios of the pairs Th/Eu, Th/Dy, and U/Th, computed at Gyr, are shown in Figure 15. The wind ejecta mass is calculated according to Eq. (1), considering either only the material directly ejected during the simulation () or including an additional 20% or 40% of the accretion disc mass ( and , respectively). Variations in both the BNS model and the assumed wind ejecta mass have a significant impact on and . This behaviour is consistent with the findings of Holmbeck et al. [42], who showed that is extremely sensitive to the ejecta and entropy. Holmbeck et al. [42] define -process stars as actinide-deficient if , actinide-normal if , and actinide-boost if . We report these three groups in Figure 15 as blue, yellow, and pink shaded regions, respectively.

Figure 15. log , log , and log computed at Gyr for all simulations listed in Table 1. From left to right, the wind ejecta mass is computed according to Eq. (1), by considering only the wind mass directly ejected during the simulation time () and by adding 20% or 40% of the accretion disc mass ( and ), respectively. The three shaded regions mark the classification scheme of -process stars as reported in Holmbeck et al. [42]: actinide-deficient (blue), actinide-normal (yellow), and actinide-boost (pink).
For the case, the Th/Dy ratios predicted by all our simulations fall in the actinide-boost region. In particular, the DD2_M10899-15570_M1 and DD2_M12980-12980_M1 simulations fall within the range [, ] observed in actinide-boost stars in the Milky Way [e.g. 42 figure 1]. When the wind contribution is larger (), DD2_M12980-12980_M1 becomes consistent with the Th/Dy ratios measured in actinide-normal metal-poor Milky Way stars, and approaches the actinide-deficient range for . The models DD2_M10899-15570_M1 and DD2_M12075-17250_M1 remain in the actinide-boost region for and start approaching the actinide-normal region for . All other models keep , consistent with actinide-boost stars, for all .
Considering the Th/Eu ratio, we find that for all models fall in the actinide-normal or actinide-boost range of the Holmbeck et al. [42] classification. Only DD2_M10899-15570_M1 is marginally consistent with the metal-poor star measurements of Roederer et al. [98]. When the wind contribution is larger (), DD2_M12980-12980_M1 and DD2_M12075-17250_M1 also become consistent with the observed stellar abundances [98].
As expected, is largely insensitive to both the BNS model and wind ejecta mass (e.g. Kullmann et al. [56]). All models predict , in good agreement with measurements in metal poor stars Roederer et al. [98].
However, we note that all of these abundance ratios are highly sensitive to the nuclear inputs adopted in the nucleosynthesis calculations (e.g. Eichler et al. [24]; Kullmann et al. [57]). This additional source of theoretical uncertainty should therefore be carefully considered when comparing nucleosynthesis predictions with observational data.
Conclusions
In this work, we have developed and publicly released a new set of -process nucleosynthesis yields based on a sample of eight BNS merger simulations in numerical relativity. To our knowledge, this is the first public release of yields and tracer trajectories obtained from full general relativity simulations of long-lived remnants with M1 neutrino transport.
Starting from a sample of eight BNS merger models spanning different EOSs, total binary masses, and mass ratios, we performed numerical relativity simulations and recorded both the thermodynamic properties of the ejecta crossing the detector surface and the thermodynamic evolution of passive tracer particles. All simulations employed the M1 neutrino transport scheme and resulted in long-lived NS remnants. We then developed a sampling algorithm to select 3000 − 5000 representative tracers per simulation and assigned a representative mass to each tracer. The resulting tracer set was post-processed with the nuclear reaction network WinNet to compute -process nucleosynthesis yields. Finally, we released the nucleosynthesis yields for all simulations, computed at day and Gyr, considering multiple prescriptions for the wind ejecta mass to account for the current uncertainties in the amount of material unbound from the remnant accretion disc. The main results of this study can be summarised as follows.
We publicly release the complete set of tracer trajectories for all BNS merger models together with the corresponding nucleosynthesis yields through the online platform MARTINI. The selected tracers accurately reproduce the thermodynamic and mass distributions of both the dynamical and wind ejecta, while the published yields at day and Gyr provide a resource for future kilonova and Galactic chemical evolution studies.
The nucleosynthesis yields exhibit a clear dependence on the merger properties. At day, while the production of light and first-peak elements is relatively robust across the simulation sample and is primarily enhanced by the neutrino-irradiated disc wind, the yields of heavy -process nuclei vary by more than one order of magnitude, with unequal-mass mergers producing the largest abundances owing to their larger fraction of neutron-rich ejecta. Compared with previous calculations employing an M0 neutrino transport scheme, our M1 simulations predict substantially larger He abundances, particularly in the wind ejecta, demonstrating the important role of neutrino irradiation in determining the ejecta composition and the resulting nucleosynthesis.
In all simulations, the total He mass fraction of the ejecta (dynamical plus wind, for ) exceeds 0.1, and in both the dynamical and wind ejecta. At day, the wind ejecta cluster around and for all our simulations with the exception of DD2_M10899-15570_M1.
At Gyr, all simulations consistently produce large and comparable yields of light, Fe-peak, and first-peak elements, while the yields of second-peak, rare-earth, and third-peak nuclei vary substantially among models, particularly in the wind ejecta, with asymmetric BNS models generally producing more rare-earth and third-peak elements. In particular, the production of first-peak elements such as Sr is remarkably robust among simulations, whereas the yields of heavier elements, including Ba, Eu, and Th, span more than one order of magnitude. These variations primarily reflect the different amounts of neutron-rich ejecta and the relative contributions of the dynamical and wind components. All models reproduce the second and third solar -process peaks, but all except DD2_M14350-14350_M1 fail to reproduce the first one, and several of them overproduce the rareearth region. The models giving the closest match to the solar abundance pattern are the equal-mass DD2_M12980-12980_M1 and SFHo_M12980-12980_M1_HR.
The nucleosynthesis yields presented in this work are, however, subject to several sources of uncertainty. From the nuclear physics side, the predicted abundances depend sensitively on the adopted nuclear mass model and reaction rates (e.g. Eichler et al. [24]). From the numerical relativity side, we have shown that the simulation resolution can influence the yields, in some cases at a level comparable to that of changing the NS EOS. In addition, the ejecta mass had not saturated at the end of our simulations, so the wind ejecta masses should be regarded as lower limits. The thermodynamic evolution of the ejecta is also influenced by the uncertainties in the modelling of the merger microphysics, and the treatment of viscosity can significantly affect the nucleosynthesis output (Fujibayashi et al. [31]; Just et al. [50]). A further source of uncertainty is the assumption of homologous expansion when post-processing tracers for the nucleosynthesis calculations (Magistrelli et al. [61]). Moreover, the present dataset is limited to BNS mergers producing long-lived remnants, and should be extended in the future to include systems undergoing prompt or delayed collapse to a BH. Finally, it will be important to assess the robustness of our results against the adopted tracer mass-assignment procedure, by comparing with state-of-the-art prescriptions in which tracers are sampled from the ejecta at the detector sphere, and their properties reconstructed by integrating their trajectories backward in time (e.g. Nishimura et al. [71]; Fujibayashi et al. [31]; Witt et al. [120]; Just et al. [50]; Jacobi et al. [46]).
Despite these uncertainties, the dataset presented here provides the first publicly available library of -process yields and tracer trajectories extracted from BNS merger simulations performed with the M1 neutrino transport scheme, offering a valuable resource for future studies of kilonovae, Galactic chemical evolution, and stellar abundances.
Data availability
The tracer trajectories (time, temperature, density, radius, and electron fraction) and the tracer masses for all the simulations presented in this work are publicly available on Zenodo at https://doi.org/10.5281/zenodo.23689687. The nucleosynthesis yields, both elemental and isotopic, at 1 day and 1 Gyr are available through the MARTINI platform at https://martini.oa-abruzzo.inaf.it.
Acknowledgements. E.L., D.V., and S.C. acknowledge funding by the European Union – NextGenerationEU RFF M4C2 1.1 PRIN 2022 project 2022JRLWHN URKA. E.L. and S.C. acknowledge founding from 1.05.24.07.02 MiniGrants RSN2 (2024). A.P. and F.M.G. are supported by the European Union under NextGenerationEU, PRIN 2022 Project No. 2022KX2Z3B. A.P. and F.M.G. also acknowledge the EuroHPC Joint Undertaking for awarding this project access to the EuroHPC supercomputer LUMI, hosted by CSC (Finland) and the LUMI consortium through a EuroHPC Extreme Scale Access call (EHPC-EXT-2022-01-046). The authors acknowledge computational resources from the INFN-Torino (Tullio server) and INFN-TIFPA (TheoServ server). The authors thank Jan Kuske, Almudena Arcones, Gabriele Cescutti, Marta Molero, and Giacomo Ricalgiano for useful interactions and comments.
Testing tracers mass assignment
We tested the tracer mass assignment procedure described in Sect. 2.3 by comparing, for each simulation, the mass distributions as functions of the electron fraction , entropy , radial velocity , and polar angle obtained directly from the mass outflow with those reconstructed from the mass-weighted tracers. The comparison was performed for every ejection time window considered in our analysis. We find an extremely good agreement between the mass distributions computed with the two different methods. In Fig. A.1 we show the comparison for the total ejecta, where the mass distributions reconstructed from the tracers (black dashed lines) closely overlap those computed from the outflow data (solid lines), demonstrating that the adopted mass assignment accurately reproduces the properties of the ejecta.

Figure A.1. Mass distribution computed from outflow (solid lines) and mass-weighted tracers (dashed black lines) as function of , , , and for all simulations reported in Table 2.
References
- [1]Abbott, B. P., Abbott, R., Abbott, T. D., et al. 2017a, Phys. Rev. Lett., 119, 161101
- [3]Antoniadis, J., Freire, P. C., Wex, N., et al. 2013, Science, 340, 448
- [4]Ardevol-Pulpillo, R., Janka, H.-T., Just, O., & Bauswein, A. 2019, MNRAS, 485, 4754
- [5]Arya, A., Damgaard, R., Sneppen, A., et al. 2026, arXiv e-prints, arXiv:2604.05812arxiv.org/abs/2604.05812
- [6]Bauswein, A., Goriely, S., & Janka, H. T. 2013, ApJ, 773, 78
- [7]Bernuzzi, S. & Hilditch, D. 2010, Phys. Rev. D, 81, 084003
- [8]Bernuzzi, S., Magistrelli, F., Jacobi, M., et al. 2025, MNRAS, 542, 256
- [9]Bezmalinovich, M. 2026, Galaxies, 14, 62
- [10]Bezmalinovich, M., Bulla, M., Gaigalas, G., et al. 2026, A&A, 708, A366
- [11]Burbidge, E. M., Burbidge, G. R., Fowler, W. A., & Hoyle, F. 1957, Reviews of Modern Physics, 29, 547
- [12]Çokluk, K. A., Perego, A., Guercilena, F. M., & Yakut, K. 2026, arXiv e-prints, arXiv:2608.16275arxiv.org/abs/2608.16275
- [13]Chiba, K., Tanaka, M., Wanajo, S., et al. 2026, arXiv e-prints, arXiv:2604.05703DOI
- [14]Cowan, J. J., Sneden, C., Beers, T. C., et al. 2005, ApJ, 627, 238
- [15]Cowan, J. J., Sneden, C., Lawler, J. E., et al. 2021, Reviews of Modern Physics, 93, 015002
- [16]Cristallo, S., Piersanti, L., Straniero, O., et al. 2011, ApJS, 197, 17
- [17]Cristallo, S., Straniero, O., Piersanti, L., & Gobrecht, D. 2015, ApJS, 219, 40
- [18]Cyburt, R. H., Amthor, A. M., Ferguson, R., et al. 2010, ApJS, 189, 240
- [20]Domoto, N., Tanaka, M., Kato, D., et al. 2022, ApJ, 939, 8
- [21]Domoto, N., Tanaka, M., Wanajo, S., & Kawaguchi, K. 2021, ApJ, 913, 26
- [22]Domoto, N., Wanajo, S., Tanaka, M., Kato, D., & Hotokezaka, K. 2025, ApJ, 978, 99
- [23]Drischler, C., Hebeler, K., & Schwenk, A. 2019, Phys. Rev. Lett., 122, 042501
- [24]Eichler, M., Sayar, W., Arcones, A., & Rauscher, T. 2019, ApJ, 879, 47
- [25]Einfeldt, B. 1988, SIAM Journal on Numerical Analysis, 25, 294
- [26]Etienne, Z., Brandt, S. R., Diener, P., et al. 2021, The Einstein Toolkit
- [27]Fahlman, S., Fernández, R., & Morsink, S. 2023, MNRAS, 526, 952
- [28]Foucart, F., Haas, R., Duez, M. D., et al. 2016, Phys. Rev. D, 93, 044019
- [29]Foucart, F., Mösta, P., Ramirez, T., et al. 2021, Phys. Rev. D, 104, 123010
- [30]Fujibayashi, S., Kiuchi, K., Nishimura, N., Sekiguchi, Y., & Shibata, M. 2018, ApJ, 860, 64
- [31]Fujibayashi, S., Wanajo, S., Kiuchi, K., et al. 2020, ApJ, 901, 122
- [32]Galeazzi, F., Kastaun, W., Rezzolla, L., & Font, J. A. 2013, Phys. Rev. D, 88, 064009
- [33]Gillanders, J. H., Sim, S. A., Smartt, S. J., Goriely, S., & Bauswein, A. 2024, MNRAS, 529, 2918
- [34]Gourgoulhon, E., Grandclément, P., Marck, J.-A., Novak, J., & Taniguchi, K. 2016, LORENE: Spectral methods differential equations solver
- [35]Gourgoulhon, E., Grandclément, P., Taniguchi, K., Marck, J.-A., & Bonazzola, S. 2001, Phys. Rev. D, 63, 064029
- [36]Harten, A., Lax, P. D., & Leer, B. v. 1983, SIAM Review, 25, 35
- [37]Hempel, M. & Schaffner-Bielich, J. 2010, Nucl. Phys. A, 837, 210
- [38]Hilditch, D., Bernuzzi, S., Thierfelder, M., et al. 2013, Phys. Rev. D, 88, 084057
- [39]Hoffman, R. D., Woosley, S. E., & Qian, Y.-Z. 1997a, Nucl. Phys. A, 621, 397
- [41]Holmbeck, E. M., Beers, T. C., Roederer, I. U., et al. 2018, ApJ, 859, L24
- [42]Holmbeck, E. M., Frebel, A., McLaughlin, G. C., et al. 2019, ApJ, 881, 5
- [43]Honda, S., Aoki, W., Ishimaru, Y., Wanajo, S., & Ryan, S. G. 2006, ApJ, 643, 1180
- [44]Horowitz, C. J. 2019, Annals of Physics, 411, 167992
- [45]Hotokezaka, K., Tanaka, M., Kato, D., & Gaigalas, G. 2023, MNRAS, 526, L155
- [46]Jacobi, M., Guercilena, F. M., Huth, S., et al. 2024, MNRAS, 527, 8812
- [47]Jacobi, M., Magistr elli, F., Loffredo, E., et al. 2026, ApJ, 999, L16
- [49]Just, O., Obergaulinger, M., & Janka, H.-T. 2015, MNRAS, 453, 3386
- [50]Just, O., Vijayan, V., Xiong, Z., et al. 2023, ApJ, 951, L12
- [51]Kawaguchi, K., Fujibayashi, S., Shibata, M., Tanaka, M., & Wanajo, S. 2021, ApJ, 913, 100
- [53]Kiuchi, K., Fujibayashi, S., Hayashi, K., et al. 2023, Phys. Rev. Lett., 131, 011401
- [54]Kodama, T. & Takahashi, K. 1975, Nucl. Phys. A, 239, 489
- [55]Koning, A., Hilaire, S., & Goriely, S. 2023, European Physical Journal A, 59, 131
- [56]Kullmann, I., Goriely, S., Just, O., et al. 2022, MNRAS, 510, 2804
- [57]Kullmann, I., Goriely, S., Just, O., Bauswein, A., & Janka, H.-T. 2023, MNRAS, 523, 2551
- [58]Lattimer, J. M. & Steiner, A. W. 2014, ApJ, 784, 123
- [59]Löffler, F., Faber, J., Bentivegna, E., et al. 2012, Classical and Quantum Gravity, 29, 15001
- [60]Longo Micchi, L. F., Magistrelli, F., Jacobi, M., & Bernuzzi, S. 2026, arXiv e-prints, arXiv:2608.10100
- [61]Magistrelli, F., Bernuzzi, S., Perego, A., Jacobi, M., & Fontes, C. J. 2026, A&A, 708, A126
- [62]Martinet, S. & Goriely, S. 2025, A&A, 694, A180
- [63]Miller, M. C., Lamb, F. K., Dittmann, A. J., et al. 2019, ApJ, 887, L24
- [64]Miller, M. C., Lamb, F. K., Dittmann, A. J., et al. 2021, ApJ, 918, L28
- [65]Möller, P., Mumpower, M. R., Kawano, T., & Myers, W. D. 2019, Atomic Data and Nuclear Data Tables, 125, 1
- [66]Möller, P., Sierk, A. J., Ichikawa, T., Iwamoto, A., & Mumpower, M. 2015, Phys. Rev. C, 91, 024310
- [67]Möller, P., Sierk, A. J., Ichikawa, T., & Sagawa, H. 2016, Atomic Data and Nuclear Data Tables, 109, 1
- [68]Mumpower, M. R., Jaffke, P., Verriere, M., & Randrup, J. 2020, Phys. Rev. C, 101, 054607
- [69]National Research Council. 2003, Connecting Quarks with the Cosmos: Eleven Science Questions for the New Century (Washington, DC: The National Academies Press)
- [70]Nedora, V., Bernuzzi, S., Radice, D., et al. 2021, ApJ, 906, 98
- [71]Nishimura, N., Takiwaki, T., & Thielemann, F.-K. 2015, ApJ, 810, 109
- [72]Oechslin, T., Janka, H.-T., & Marek, A. 2007, A&A, 467, 395
- [73]Oertel, M., Hempel, M., Klähn, T., & Typel, S. 2017, Reviews of Modern Physics, 89, 015007
- [74]Panov, I. V., Kolbe, E., Pfeiffer, B., et al. 2005, Nucl. Phys. A, 747, 633
- [75]Panov, I. V., Korneev, I. Y., Rauscher, T., et al. 2010, A&A, 513, A61
- [76]Perego, A., Bernuzzi, S., & Radice, D. 2019, European Physical Journal A, 55, 124
- [77]Perego, A., Thielemann, F. K., & Cescutti, G. 2021, in Handbook of Gravitational Wave Astronomy, 13
- [78]Perego, A., Vescovi, D., Fiore, A., et al. 2022, ApJ, 925, 22
- [79]Pian, E., D’Avanzo, P., Benetti, S., et al. 2017, Nature, 551, 67
- [80]Prantzos, N., Abia, C., Cristallo, S., Limongi, M., & Chieffi, A. 2020, MNRAS, 491, 1832
- [81]Raaijmakers, G., Greif, S. K., Hebeler, K., et al. 2021, ApJ, 918, L29
- [82]Radice, D. & Bernuzzi, S. 2023, ApJ, 959, 46
- [83]Radice, D. & Bernuzzi, S. 2024, in Journal of Physics Conference Series, Vol. 2742, Journal of Physics Conference Series (IOP), 012009
- [84]Radice, D., Bernuzzi, S., & Perego, A. 2020, Annual Review of Nuclear and Particle Science, 70, 95
- [85]Radice, D., Bernuzzi, S., Perego, A., & Haas, R. 2022, MNRAS, 512, 1499
- [86]Radice, D., Perego, A., Hotokezaka, K., et al. 2018, ApJ, 869, 130
- [87]Radice, D. & Rezzolla, L. 2012, A&A, 547, A26
- [88]Radice, D., Rezzolla, L., & Galeazzi, F. 2014a, in APS Meeting Abstracts, Vol. 2014, APS April Meeting Abstracts, R15.003
- [91]Rahmouni, S., Tanaka, M., Domoto, N., et al. 2025, ApJ, 980, 43
- [92]Reichert, M., Winteler, C., Korobkin, O., et al. 2023, ApJS, 268, 66
- [94]Reisswig, C., Ott, C. D., Abdikamalov, E., et al. 2013b, Phys. Rev. Lett., 111, 151101
- [95]Ricigliano, G., Jacobi, M., & Arcones, A. 2024, MNRAS, 533, 2096
- [97]Rochman, D., Koning, A., Goriely, S., & Hilaire, S. 2025, Nucl. Phys. A, 1053, 122951
- [98]Roederer, I. U., Kratz, K.-L., Frebel, A., et al. 2009, ApJ, 698, 1963
- [99]Roederer, I. U., Lawler, J. E., Sobeck, J. S., et al. 2012, ApJS, 203, 27
- [100]Rosswog, S., Sarin, N., Nakar, E., & Diener, P. 2025, MNRAS, 538, 907
- [101]Rosswog, S., Sollerman, J., Feindt, U., et al. 2018, A&A, 615, A132
- [102]Schnetter, E., Hawley, S. H., & Hawke, I. 2004, Classical and Quantum Gravity, 21, 1465
- [103]Shibata, M. & Hotokezaka, K. 2019, Annual Review of Nuclear and Particle Science, 69, 41
- [104]Shlomo, S., Kolomietz, V. M., & Colò, G. 2006, European Physical Journal A, 30, 23
- [105]Siegel, D. M. & Metzger, B. D. 2017, Phys. Rev. Lett., 119, 231102
- [106]Smartt, S. J., Chen, T. W., Jerkstrand, A., et al. 2017, Nature, 551, 75
- [107]Sneden, C., Cowan, J. J., & Gallino, R. 2008, ARA&A, 46, 241
- [108]Sneppen, A., Damgaard, R., Watson, D., et al. 2024, A&A, 692, A134
- [109]Sneppen, A., Just, O., Bauswein, A., et al. 2026, Phys. Rev. D, 113, 063038
- [110]Sneppen, A. & Watson, D. 2023, A&A, 675, A194
- [111]Sprouse, T. M., Lund, K. A., Miller, J. M., McLaughlin, G. C., & Mumpower, M. R. 2024, Astrophys. J., 962, 79
- [112]Steiner, A. W., Prakash, M., Lattimer, J. M., & Ellis, P. J. 2005, Phys. Rep., 411, 325
- [113]Suresh, A. & Huynh, H. T. 1997, Journal of Computational Physics, 136, 83
- [114]Thielemann, F. K., Eichler, M., Panov, I. V., & Wehmeyer, B. 2017, Annual Review of Nuclear and Particle Science, 67, 253
- [115]Typel, S., Röpke, G., Klähn, T., Blaschke, D., & Wolter, H. H. 2010, Phys. Rev. C, 81, 015803
- [116]Vincent, T., Foucart, F., Duez, M. D., et al. 2020, Phys. Rev. D, 101, 044053
- [117]Wanajo, S. & Ishimaru, Y. 2006, Nucl. Phys. A, 777, 676
- [118]Wang, M., Huang, W. J., Kondev, F. G., Audi, G., & Naimi, S. 2021, Chinese Physics C, 45, 030003
- [119]Watson, D., Hansen, C. J., Selsing, J., et al. 2019, Nature, 574, 497
- [120]Witt, M., Psaltis, A., Yasin, H., et al. 2021, ApJ, 921, 19
- [121]Xiong, Z., Martinez-Pinedo, G., Just, O., & Sieverding, A. 2024, Phys. Rev. Lett., 132, 192701
- [122]Zappa, F., Bernuzzi, S., Radice, D., & Perego, A. 2023, MNRAS, 520, 1481