betaFeedback

WorksPhysics and Astronomy

What constrains the neutron-star crust? Present data, future measurements, and EOS doppelgängers

Abstract

The neutron-star crust plays a central role in pulsar glitches, thermal evolution, and stellar oscillations, yet its properties remain difficult to constrain. We use Bayesian inference with a unified crust-core equation of state to assess what present nuclear and astrophysical data tell us about the crust and what future stellar measurements can add. Within this model, present crust constraints are dominated by chiral effective field theory, with laboratory data providing additional information and current stellar observations having a smaller effect. The adopted theoretical bands determine how tightly the crust properties are constrained, while the treatment of neutron-rich surface energy affects their inferred values. Forecasts centred on a single reference equation of state show that more precise measurements of radius, tidal deformability, and total moment of inertia mainly constrain crust thickness. They provide little additional information about the crust-core transition, crust mass, or crustal moment of inertia and do not recover the precision lost when the chiral effective field theory constraint is removed. Two models with nearly identical global stellar observables can have crust masses and crustal moments of inertia differing by about a factor of two, yet remain indistinguishable at the assumed precision. Precise global measurements therefore need not uniquely determine the crust. Observations that directly probe the crust are needed to distinguish these models.

Introduction

The neutron-star (NS) crust contains only a small fraction of the stellar mass and radius, yet it substantially influences a wide range of observational phenomena. Its properties are central to models of pulsar glitches [2, 45, 3, 13], cooling of transiently accreting stars [7, 55], magnetar oscillations and crust failure [39, 74, 34, 58], pre-merger crust shattering, and the ejection of neutron-rich matter [77, 29]. Thermal and magnetic transport through the crust also helps set the observed X-ray surface temperature [14, 28]. Despite its importance for NS observables, the composition, microscopic structure, and extent of the crust remain difficult to constrain. Although laboratory nuclei constrain matter at sub-saturation densities, the extreme neutron excess, nuclear lattice, and unbound neutrons of the inner crust remain difficult to study experimentally and theoretically [14]. Most observational signals sensitive to the crust do not isolate its contribution. Their interpretation depends on its coupling to the core, superfluid components, magnetic field, and thermal structure [14, 3, 13, 55].

Over the past decade, nuclear theory, laboratory experiments, and multimessenger observations have substantially improved our knowledge of the NS equation of state (EOS) [23, 4, 61, 56, 37, 76]. Nuclear masses and neutron skins probe neutron-rich matter below saturation density, microscopic many-body calculations constrain matter up to about saturation density, and heavy-ion collisions extend the laboratory reach to higher densities. Mass measurements of massive pulsars, mass–radius measurements from the Neutron Star Interior Composition Explorer (NICER), and tidal information from binary mergers provide complementary constraints on cold matter in NS cores. These inputs have been combined in a wide range of EOS studies based on different descriptions of dense matter and assumptions about its composition and phase structure [26, 4, 76, 37]. Many such studies focus on the core and attach a standard crust EOS to each sampled core EOS. This approximation has little effect on the maximum mass and only a modest effect on other global properties. It can nevertheless change stellar radii at the few-percent level, especially at low mass, and produce much larger errors in the crust thickness [26, 25, 57]. A consistent crust–core description is particularly important for estimating the radii of low-mass neutron stars and the relation between their baryonic and gravitational masses [59].

A fixed crust may therefore be adequate for many studies of the core EOS, but it cannot be used to determine what constrains the crust itself. The crust–core transition is set below saturation density by the competition between the bulk energy of neutron-rich matter and the surface and Coulomb energies of nuclear clusters. To leading order, the transition pressure sets the crust mass and crustal moment of inertia, while the difference in baryon chemical potential between the stellar surface and the crust–core boundary determines the crust thickness (Carreau et al. 2019a,b; Newton et al. 2022). At the same time, these stellar crust properties depend on the mass, radius, and therefore the EOS of the whole star. A crust calculation must consequently be embedded in a complete crust–core EOS, on which nuclear and astrophysical constraints act together. Moreover, the part of the EOS tested by a global observable depends on the stellar mass. Weih et al. [80], for example, showed that measurements at different masses do not constrain the same density region, and that intermediate-mass stars can provide greater leverage on the lower-density EOS than the most massive stars.

Several Bayesian studies have treated the crust and core consistently or have focused specifically on crust properties (Carreau et al. 2019a,b; Balliet et al. 2021; Dinh Thi et al. 2021; Newton et al. 2022; Scurto et al. 2024; Burrello et al. 2025; Parmar & Bombaci 2026a). These studies show that microscopic neutron-matter calculations and the poorly known surface energy of very neutron-rich clusters strongly affect the crust–core transition [10 b]. Nuclear masses, neutron-skin measurements, and finite-size terms further constrain the transition and the pasta layers [6, 21, 53]. However, the relative contributions of individual nuclear and astrophysical constraints, and the extent to which future stellar measurements can replace low-density nuclear information, remain less systematically explored.

Here, we address these questions using Bayesian inference within a unified crust–core model. We compare combinations of present constraints, variations in the nuclear model, and forecasts across stellar masses and measurement precisions. We trace how the inferred crust–core transition, crust thickness, crust mass, and crustal moment of inertia respond when individual nuclear and astrophysical constraints are removed. We focus in particular on chiral effective field theory (χEFT) because previous Bayesian studies have found microscopic neutron-matter calculations to be among the strongest constraints on the crust–core transition and related crust properties (Carreau et al. 2019a,b; Newton et al. 2022; Burrello et al. 2025). We then examine whether precise measurements of radius, tidal deformability, and total moment of inertia can provide comparable information across different stellar masses. Finally, we test whether precise global measurements can distinguish different crusts. We identify two unified EOSs with almost identical radii, tidal deformabilities, and maximum masses, but with well-separated crust properties. We then generate mock radius, tidal deformability, and moment-of-inertia measurements from each EOS, using the same uncertainties, and infer the crust in the full model. These “crust dopplegängers” extend the EOS-degeneracy studies of Raithel & Most (2023a,b) to the crust itself.

Unified crust and core model

We describe the inner crust with the compressible liquid-drop model (CLDM) [10 b]. In the CLDM, matter is described within a spherical Wigner–Seitz cell containing a neutron-rich nuclear cluster, electrons, and unbound neutrons. At each mean density, the equilibrium composition is found by minimizing the total energy, including bulk, surface, curvature, and Coulomb contributions. The competition between these contributions determines whether clustered or uniform matter is favoured and thus sets the crust–core transition. For the outer crust, we use experimental nuclear masses where available. In the deeper layers, where increasing neutron-rich nuclei appear and measured masses are unavailable, we use the HFB-24 mass model [30]. We keep this outer-crust EOS fixed for all sampled EOSs.

The bulk energies of the cluster, neutron gas, and uniform core are supplied by the empirical metamodel of Margueron et al. [46]. The interaction contribution to the energy per baryon is expanded to fourth order around saturation density. A low-density correction ensures that the energy per baryon vanishes at zero density. The parameter vector θ\theta contains thirteen empirical parameters,

θ={nsat,Esat,Ksat,Qsat,Zsat,Jsym,Lsym,Ksym,Qsym,Zsym,m∗/m,Δm∗,b}.\theta= \{n_{\mathrm{sat}}, E_{\mathrm{sat}}, K_{\mathrm{sat}}, Q_{\mathrm{sat}}, Z_{\mathrm{sat}}, J_{\mathrm{sym}}, L_{\mathrm{sym}}, K_{\mathrm{sym}}, Q_{\mathrm{sym}}, Z_{\mathrm{sym}}, m^{*}/m, \Delta m^{*}, b\}.

Here, nsatn_{\mathrm{sat}} is the saturation density of symmetric nuclear matter, and EsatE_{\mathrm{sat}} is its energy per baryon at saturation. The coefficients KsatK_{\mathrm{sat}}, QsatQ_{\mathrm{sat}}, and ZsatZ_{\mathrm{sat}} are the incompressibility, skewness, and kurtosis parameters, respectively, and characterize the second-, third-, and fourth-order density dependence of the energy per baryon around nsatn_{\mathrm{sat}}. In the isovector sector, JsymJ_{\mathrm{sym}} is the symmetry energy at saturation, while LsymL_{\mathrm{sym}}, KsymK_{\mathrm{sym}}, QsymQ_{\mathrm{sym}}, and ZsymZ_{\mathrm{sym}} are its slope, curvature, skewness, and kurtosis parameters, respectively, evaluated at nsatn_{\mathrm{sat}}. The ratio m∗/mm^{*}/m is the nucleon effective mass in symmetric matter at saturation divided by the bare nucleon mass mm. The dimensionless splitting Δm∗=(mn∗−mp∗)/m\Delta m^{*}=(m_n^{*}-m_p^{*})/m is evaluated in pure-neutron matter at saturation, where mn∗m_n^{*} and mp∗m_p^{*} are the neutron and proton effective masses. These two parameters control the momentum-dependent kinetic energy, while the dimensionless parameter bb controls the low-density correction. In benchmarks against existing nuclear models, Margueron et al. [46] showed that the fourth-order metamodel can reproduce a broad range of smooth nucleonic EOSs when its high-order coefficients are adjusted to their density dependence. Bayesian calculations with relativistic mean-field models and nonrelativistic metamodels have yielded similar crust-property distributions when low-density nuclear constraints are included [73, 57]. We use the metamodel here because its computational efficiency makes the repeated inference across constraint combinations, model variants, and forecast scenarios practical. The independent uniform priors adopted for these parameters are listed in Table B.1 of Appendix § 8.§ 1.

For each sampled EOS, the surface and curvature coefficients are fitted anew to the masses from the 2020 Atomic Mass Evaluation (AME2020) [79]. This keeps the surface and bulk contributions consistent for each EOS. We adopt the surface-tension parametrization introduced by Ravenhall et al. [64]. It is given in Eq. (A.4) of Appendix § 7, where the exponent pp controls the extrapolation to small proton fractions. We adopt p=3p = 3, a commonly used reference choice in liquid-drop crust models [52, 10, 32]. This choice reproduces dynamical-spinodal transition densities reasonably well for several nuclear interactions [10]. Comparisons with extended Thomas–Fermi calculations also favour values close to three, although the preferred value depends on the nuclear interaction and treatment of the surface [32]. We therefore repeat the calculation with p=2.5p = 2.5 and 3.53.5 to assess this model dependence.

The nonuniform crust and uniform core use the same nuclear interaction. The crust–core transition occurs where uniform matter becomes energetically favoured over the optimized nuclear configuration. We retain only EOSs that are stable, thermodynamically consistent, and causal. A one-sided mass constraint suppresses EOSs whose maximum mass Mmax⁡M_{\max} falls below 2 M⊙2\,M_{\odot}, where M⊙M_{\odot} is the solar mass (Appendix B). We collect the five crust quantities studied here in the vector ycrust\mathbf{y}_{\mathrm{crust}},

ycrust={nt,Pt,ΔRcrust,Mcrust,Icrust/I},(1)\mathbf{y}_{\mathrm{crust}} = \{n_{\mathrm{t}}, P_{\mathrm{t}}, \Delta R_{\mathrm{crust}}, M_{\mathrm{crust}}, I_{\mathrm{crust}}/I\}, \tag*{(1)}

where ntn_{\mathrm{t}} and PtP_{\mathrm{t}} are the baryon density and pressure at the crust–core transition. The crust thickness is ΔRcrust=R−Rcore\Delta R_{\mathrm{crust}} = R - R_{\mathrm{core}}, and its gravitational mass is Mcrust=M−McoreM_{\mathrm{crust}} = M - M_{\mathrm{core}}, where MM and RR are the stellar gravitational mass and radius, and McoreM_{\mathrm{core}} and RcoreR_{\mathrm{core}} are their values at the transition. The ratio Icrust/II_{\mathrm{crust}}/I is the crustal contribution IcrustI_{\mathrm{crust}} to the total stellar moment of inertia II. These three stellar crust quantities depend on MM. We evaluate them at 1.4 M⊙1.4\,M_{\odot} unless stated otherwise. Mass arguments such as (1.4) denote stellar masses in units of M⊙M_{\odot}. These quantities describe the crust–core boundary and the crust contributions to the stellar radius, mass, and moment of inertia. The metamodel expansion and CLDM energy contributions are summarized in Appendix A. The priors and likelihoods are listed in Appendix B.

Present constraints and forecast construction

Present nuclear and astrophysical constraints

The present constraints comprise nuclear-matter calculations, laboratory measurements, and neutron-star observations, which probe different density ranges and EOS properties. The combined likelihood is written as

Lpresent=LphysLsatLlabLχEFTLMmax⁡LNICERLGW170817.(2)\mathcal{L}_{\mathrm{present}} = \mathcal{L}_{\mathrm{phys}}\mathcal{L}_{\mathrm{sat}}\mathcal{L}_{\mathrm{lab}}\mathcal{L}_{\chi\mathrm{EFT}}\mathcal{L}_{M_{\max}}\mathcal{L}_{\mathrm{NICER}}\mathcal{L}_{\mathrm{GW170817}}. \tag*{(2)}

The physical filter Lphys\mathcal{L}_{\mathrm{phys}} enforces thermodynamic consistency, stability, positive symmetry energy, and causality throughout the unified EOS. The saturation term Lsat\mathcal{L}_{\mathrm{sat}} constrains the saturation density, energy per baryon, and incompressibility of symmetric nuclear matter [8, 12, 40, 68, 46]. Additional laboratory constraints at finite density enter through Llab\mathcal{L}_{\mathrm{lab}} and constrain the symmetry energy, its density dependence, and the pressure of symmetric matter. These inputs are drawn from the experimental compilation of Tsang et al. [76]. The individual measurements and their original sources are listed in Table B.2.

The low-density EOS is constrained by χEFT\chi\mathrm{EFT} calculations of symmetric and pure-neutron matter between 0.02 and 0.20 fm−3\mathrm{fm}^{-3} [22, 33]. This interval covers densities relevant to the inner crust and the transition to uniform matter, making χEFT\chi\mathrm{EFT} the principal microscopic constraint in this regime. The factor LχEFT\mathcal{L}_{\chi\mathrm{EFT}} assigns equal weight to EOSs whose energies lie within the χEFT\chi\mathrm{EFT} bands at the evaluated densities in this interval and rejects those outside. In the reference calculation, the band boundaries are widened by 5%. Sensitivity to this band-widening prescription is assessed using the original bands and widenings of 10% and 20%. A separate assessment of regulator-cutoff dependence and order-by-order χEFT\chi\mathrm{EFT} truncation uncertainties is beyond the scope of this work [see 31 for a quantified truncation-error treatment]. The percentage widenings mimic a broader theoretical uncertainty band and test its effect on the inferred crust properties, but do not provide a statistical estimate of these uncertainties. Calculations without the χEFT\chi\mathrm{EFT} constraint assess the independent constraining power of laboratory and stellar data.

The astrophysical factors constrain the global stellar structure. The term LMmax⁡\mathcal{L}_{M_{\max}} constrains the maximum supported gravitational mass Mmax⁡M_{\max} through a smooth penalty below 2 M⊙2\,M_{\odot}, motivated by the observed massive neutron stars [5, 18]. The NICER term includes the mass–radius constraints for PSRs J0030+0451 [66], J0437−4715 [15], J0614−3320 [49], and J0740+6620 [67]. For each source, the measured mass–radius distribution is evaluated along the predicted R(M)R(M) curve, retaining the uncertainty in stellar mass. The GW170817 term uses the dimensionless binary tidal deformability Λ~\widetilde{\Lambda} inferred by Abbott et al. [1], with the chirp mass fixed and the mass ratio marginalized. This quantity combines the individual dimensionless tidal deformabilities Λ(M)\Lambda(M) according to Eq. (B.5). The adopted constraints, their sources, and the likelihood definitions are given in Appendix B.1.

Forecasts of future measurements

Measurements of stellar radius, tidal deformability, and moment of inertia probe the EOS through the global structure of the star. We assess how much improved measurements of these observables can reduce crust uncertainties beyond the present constraints, and whether this information depends on the low-density χEFT\chi\mathrm{EFT} input.

We construct mock measurements using a single reference EOS, TREF\mathrm{T}_{\mathrm{REF}}, selected from the full present-data posterior near the nuclear-parameter medians. For each observable and stellar mass, we take this EOS prediction as the central measured value and assign the uncertainty specified below. For example, a 2% radius forecast is centred on the predicted radius and has a standard deviation equal to 2% of that value. We then combine the mock measurements with the present constraints to infer the crust properties.

The central values are not randomly displaced by measurement noise. These zero-noise, or Asimov, data [16] allow comparisons of measurement precision and combinations of observables for the same assumed underlying EOS. The forecast gains are conditional on this reference EOS and are not averaged over all EOSs allowed by present data. Systematic errors from instrument calibration or from approximations in the gravitational-wave models used to infer tidal deformability are not included. Appendix B.2 gives further details.

The radius forecasts span 1.2–1.8 M⊙M_{\odot} in steps of 0.2 M⊙M_{\odot}, with 2% uncertainties in both mass and radius, following the precision benchmark used by Huang et al. [36] for future X-ray measurements. Adopting the same fractional precision at each mass isolates the mass dependence of the constraints. We label a binary forecast by Λ~m1+m2\widetilde{\Lambda}_{m_1+m_2}, with the component masses in solar units. Tidal forecasts assume a 10% uncertainty for equal-mass binaries with component masses of 1.2, 1.4, and 1.6 M⊙M_{\odot}. We also test 5% and 20% uncertainties for the 1.4 + 1.4 M⊙M_{\odot} binary. These precisions are motivated by prospects for next-generation gravitational-wave detectors [38], but serve as benchmarks rather than predictions for a specific event. For the moment of inertia of PSR J0737−3039A, with mass 1.338 M⊙M_{\odot}, we adopt the 10% precision considered by Lim et al. [43] and a more optimistic 5% case. Combined forecasts test the complementarity of the three observables. The “realistic combined” case uses a 2% mass–radius measurement at 1.8 M⊙M_{\odot}, a 10% tidal measurement for a 1.4 + 1.4 M⊙M_{\odot} binary, and a 10% measurement of I(1.338 M⊙)I(1.338\,M_{\odot}). The “multi-star combined” case replaces the single mass–radius measurement with five at 1.2, 1.4, 1.6, 1.8, and 2.0 M⊙M_{\odot}, each at the same 2% precision. We also consider these five mass–radius measurements alone.

Selected forecasts are repeated without the χ\chiEFT constraint, keeping the reference EOS and forecast measurements fixed. This comparison isolates the information supplied by future observations independently of the low-density χ\chiEFT input.

Statistical analysis

For each crust quantity XX, we denote its weighted posterior ath percentile by qa(X)q_a(X) and use it to define the central widths W68=q84−q16W_{68}=q_{84}-q_{16} and W90=q95−q5W_{90}=q_{95}-q_{5}. Present-data comparisons use W68case/W68refW_{68}^{\mathrm{case}}/W_{68}^{\mathrm{ref}}, with values above unity indicating broader intervals. Forecasts use the inverse ratio,

GX=W68ref(X)W68case(X).(3)G_X=\frac{W_{68}^{\mathrm{ref}}(X)}{W_{68}^{\mathrm{case}}(X)}. \tag*{(3)}

Values of GXG_X above unity indicate narrower credible intervals, while values below unity indicate broader intervals relative to the reference case. The geometric mean, Ggeom=(∏XGX)1/5G_{\mathrm{geom}}=\left(\prod_X G_X\right)^{1/5}, summarizes the width changes across the five crust quantities. Changes in posterior location and shape are additionally quantified by the marginal Kullback–Leibler divergence DKLXD_{\mathrm{KL}}^X (Kullback & Leibler [42]),

DKLX(case∥ref)=∫pcase(X)log⁡2[pcase(X)pref(X)] dX.(4)D_{\mathrm{KL}}^X(\mathrm{case}\Vert\mathrm{ref})=\int p_{\mathrm{case}}(X)\log_2\left[\frac{p_{\mathrm{case}}(X)}{p_{\mathrm{ref}}(X)}\right]\,\mathrm{d}X. \tag*{(4)}

Here pcase(X)p_{\mathrm{case}}(X) and pref(X)p_{\mathrm{ref}}(X) are the normalized marginal posterior densities for the case and reference calculations. The base-two logarithm expresses DKLXD_{\mathrm{KL}}^X in bits. It is zero for identical distributions and positive otherwise. It is not symmetric under exchange of the two distributions and is not a detection significance.

For forecasts without χ\chiEFT, we also define

GXobs=W68no χEFT(X)W68no χEFT+future(X),RXcur=W68no χEFT+future(X)W68present(X),RXmatch=W68no χEFT+future(X)W68present+future(X).(5)\begin{aligned} G_X^{\mathrm{obs}}&=\frac{W_{68}^{\mathrm{no\ }\chi\mathrm{EFT}}(X)}{W_{68}^{\mathrm{no\ }\chi\mathrm{EFT+future}}(X)},\\ R_X^{\mathrm{cur}}&=\frac{W_{68}^{\mathrm{no\ }\chi\mathrm{EFT+future}}(X)}{W_{68}^{\mathrm{present}}(X)},\qquad R_X^{\mathrm{match}}=\frac{W_{68}^{\mathrm{no\ }\chi\mathrm{EFT+future}}(X)}{W_{68}^{\mathrm{present+future}}(X)}. \tag*{(5)} \end{aligned}

Here GXobsG_X^{\mathrm{obs}} measures narrowing relative to the present-data posterior without χ\chiEFT. The ratios RXcurR_X^{\mathrm{cur}} and RXmatchR_X^{\mathrm{match}} compare the resulting width with the full present-data posterior and the corresponding χ\chiEFT-conditioned forecast, respectively. Unity therefore indicates recovery of the present precision for RXcurR_X^{\mathrm{cur}}, or equal forecast widths with and without χ\chiEFT for RXmatchR_X^{\mathrm{match}}.

Finally, we select pairs from the full present-data posterior with similar R(M)R(M), Λ(M)\Lambda(M), and maximum mass but different crust properties. Moment of inertia is excluded from the selection to retain it as an independent test. Forecast recoveries centred on each member test whether future observations can distinguish these crusts. This crust-doppelgänger test is motivated by the analysis of Raithel & Most (2023a,b). The selection and recovery criteria are defined in Sect. 5.5, with supporting details in Appendix C.

These tests measure distinguishability at the assumed precision in observable space. Detector-specific claims would additionally require waveform and noise modeling, source populations, and selection effects ([35, 54, 38]).

Results and discussion

Present constraints on crust properties

We compare thirteen configurations of the constraints and nuclear model to assess how nuclear and stellar data constrain the crust. The physical baseline includes the admissibility conditions and the lower bound on maximum mass. The nuclear-only case adds saturation, laboratory, and χ\chiEFT constraints, while the astrophysical-only case adds NICER and GW170817. Figure 1 shows the corresponding distributions of the five crust quantities. With all present constraints included, their medians and central 68% credible intervals are

Figure 1. Present constraints on the five crust quantities. The first group of rows compares the physical baseline with the nuclear-only, astrophysical-only, and full present-data calculations. The remaining groups show the effects of removing individual constraints, varying the χ\chiEFT envelope, and changing the surface exponent. The shaded shapes show marginalized probability densities. Thin and thick horizontal segments mark the central 90% and 68% credible intervals, respectively. Open circles mark the medians.

nt=0.0811−0.0069+0.0084 fm−3,Pt=0.396−0.068+0.078 MeV fm−3,ΔRcrust(1.4)=1.048−0.062+0.063 km,Mcrust(1.4)=0.0243−0.0039+0.0041 M⊙,(Icrust/I)(1.4)=0.0355−0.0054+0.0057.(6)\begin{aligned} n_{\mathrm{t}} &= 0.0811^{+0.0084}_{-0.0069}\ \mathrm{fm}^{-3},\\ P_{\mathrm{t}} &= 0.396^{+0.078}_{-0.068}\ \mathrm{MeV\,fm}^{-3},\\ \Delta R_{\mathrm{crust}}(1.4) &= 1.048^{+0.063}_{-0.062}\ \mathrm{km},\\ M_{\mathrm{crust}}(1.4) &= 0.0243^{+0.0041}_{-0.0039}\ M_\odot,\\ (I_{\mathrm{crust}}/I)(1.4) &= 0.0355^{+0.0057}_{-0.0054}. \tag*{(6)} \end{aligned}

Table D.1 also lists the nuclear-parameter posteriors and the central 90% intervals. These intervals describe the uncertainty within the adopted nuclear model and constraints. Nuclear constraints dominate the present inference: relative to the physical baseline, laboratory data together with χ\chiEFT give width ratios of 0.318–0.365. Including stellar observations gives ratios of 0.305–0.436. Stellar data alone narrow only the crust thickness, to 0.888 times its baseline width. The other four central intervals become broader as the posterior shapes change. Without nuclear information, current stellar observations therefore provide limited constraints on the transition, crust mass, and crustal moment of inertia.

Figure 2 summarizes the changes in widths and medians. Without χ\chiEFT, the five widths increase by factors of 2.08–2.96, because a wider range of neutron-rich matter properties is then allowed near the crust–core transition. Removing both saturation and finite-density laboratory constraints increases them by factors of 1.11–1.27, whereas omitting NICER or GW170817 gives most width ratios between 0.95 and 1.05. Some transition intervals become narrower without NICER because removing NICER changes the shape of the posterior. This change does not by itself indicate a stronger constraint. The inferred precision is also sensitive to the adopted χ\chiEFT band width. Relative to the reference 5% widening, the original band gives width ratios of 0.42–0.78. Widenings of 10% and 20% give ratios of 1.22–1.38 and 1.49–1.97, respectively, with smaller changes in the medians.

Figure 2. Sensitivity of inferred crust properties to present constraints and nuclear modelling. All cases are compared with the full present-data calculation (“All present”), which uses a 5% χ\chiEFT-band widening and p=3p = 3. Left: ratio of the central 68% credible-interval width to its reference value. Numbers give the ratios, while colours show their base-two logarithms. Red indicates broader intervals and blue narrower intervals. A value of unity indicates no change in width. Right: change in the median divided by the reference 68% width. Positive (red) and negative (blue) values indicate shifts to higher and lower values, respectively. The first group of rows compares the physical, nuclear, and astrophysical constraints. The remaining groups show the effects of omitting individual inputs, varying the χ\chiEFT-band widening, and changing the surface exponent pp. Crust thickness, mass, and fractional moment of inertia are evaluated at 1.4 M⊙M_\odot.

Changing the neutron-rich surface energy shifts the inferred crust properties while leaving their credible widths nearly unchanged. Relative to p=3p = 3, setting p=2.5p = 2.5 lowers the medians by 0.32–0.52 reference widths, while p=3.5p = 3.5 raises them by 0.17–0.32 widths. The separation between the two choices reaches 0.49–0.84 reference widths, although the central 68% width ratios remain close to unity (Table D.2). This sensitivity to the neutron-rich surface energy is consistent with earlier crust calculations. Newton et al. [52] showed that the surface energy at small proton fractions strongly affects the inner-crust composition and transition density. Carreau et al. [10] subsequently found that increasing pp raises both the transition density and pressure, as in our results, and that uncertainty in pp broadens their inferred ranges. The importance of treating the surface and bulk contributions consistently was also demonstrated by Dinh Thi et al. [21]. The shifts between surface prescriptions are a substantial fraction of the present uncertainty, which is not captured by the intervals at fixed pp. The three choices of pp test this sensitivity without providing a full uncertainty estimate for the surface functional. Joint inference of bulk and surface parameters offers a way to include this additional uncertainty ([41]).

The same trends are seen across stellar masses from 1.0 to 2.0 M⊙M_\odot, as shown in Figure 3. The corresponding crust properties are listed in Table D.3. Over this mass range, removing χ\chiEFT increases the widths by factors of 1.81–3.14, while removing laboratory constraints gives factors of 1.08–1.30. The width ratios remain close to unity when NICER or GW170817 is omitted (about 0.93–1.07) or the surface model is changed (about 0.92–1.08).

Figure 3. Mass dependence of the central 68% widths for crust thickness (top), crust mass (middle), and fractional crustal moment of inertia (bottom). Each curve is divided by the full present-data width at the same stellar mass. The columns show the information build-up, leave-one-out tests, χ\chiEFT-envelope changes, and surface-energy models over 1.0–2.0 M⊙M_\odot.

The inferred transition agrees with earlier unified calculations. Carreau et al. [10] obtained nt=0.072±0.011 fm−3n_{\mathrm{t}} = 0.072 \pm0.011\ \mathrm{fm}^{-3} and Pt=0.339±0.115 MeV fm−3P_{\mathrm{t}} = 0.339 \pm0.115\ \mathrm{MeV}\ \mathrm{fm}^{-3}, while Newton et al. [53] found Pt=0.33±0.07 MeV fm−3P_{\mathrm{t}} = 0.33 \pm0.07\ \mathrm{MeV}\ \mathrm{fm}^{-3} from neutron skins and χ\chiEFT neutron matter. Burrello et al. [9] obtained nt=0.074±0.014 fm−3n_{\mathrm{t}} = 0.074 \pm0.014\ \mathrm{fm}^{-3}, while Göttling et al. [31] found 0.062–0.088 fm−3\mathrm{fm}^{-3}. These studies use different surface models and treatments of theoretical uncertainty. They nevertheless favour a similar transition-density range, while their pressure estimates vary more. The leading contribution from low-density theory and the additional constraint from laboratory data are also consistent with previous work ([10, 53, 9]). Within the adopted unified model, the inferred crust properties are mainly constrained by nuclear information and remain only weakly sensitive to the inclusion of current NICER and GW170817 data.

Constraints from future stellar measurements

We first examine the improvement obtained by adding each forecast measurement individually to the full present-data constraints, using the scenarios defined in Sect. 3.2.

Crust thickness shows the largest improvement among the five crust quantities. Its strongest constraint comes from the optimistic 5% measurement of the binary tidal deformability Λ~\tilde{\Lambda} for a 1.4 + 1.4 M⊙M_\odot system, which reduces the central 68% credible-interval width by about 21%. Table D.4 lists the width gains and information gains for all forecast cases. In Figure 4, individual 2% mass–radius measurements give thickness gains of 1.09–1.12. The other width ratios remain within about 0.96–1.04, apart from a gain of 1.05 in ntn_{\mathrm{t}} at 1.2 M⊙M_\odot. The geometric-mean gain is 1.00–1.01, with no clear advantage for any mass between 1.2 and 1.8 M⊙M_\odot.

Figure 4. Change in the central 68% width after each idealized future measurement. The first fourteen rows show forecasts that retain χEFT\chi_{\mathrm{EFT}}, with GX=W68present/W68present+futureG_X = W^{\mathrm{present}}_{68}/W^{\mathrm{present+future}}_{68}. The last four rows show forecasts without χEFT\chi_{\mathrm{EFT}}. In each of these cells, the upper number is GXobs=W68no χEFT/W68no χEFT+futureG^{\mathrm{obs}}_X = W^{\mathrm{no\ }\chi_{\mathrm{EFT}}}_{68}/W^{\mathrm{no\ }\chi_{\mathrm{EFT}}+\mathrm{future}}_{68} and the lower number is RX=W68no χEFT+future/W68presentR_X = W^{\mathrm{no\ }\chi_{\mathrm{EFT}}+\mathrm{future}}_{68}/W^{\mathrm{present}}_{68}. Blue cells indicate narrower intervals and red cells broader intervals. In the last four rows, the colour represents GXobsG^{\mathrm{obs}}_X only. The final column is the geometric mean over the five crust quantities.

For the 1.4 + 1.4 M⊙M_\odot binary, relaxing the tidal uncertainty from 5% to 10% and 20% reduces the thickness gain from 1.27 to 1.17 and 1.06, respectively. The latter two correspond to width reductions of about 15% and 6%. The other four width ratios remain within about 0.97–1.03. At fixed 10% tidal uncertainty, the thickness gain increases from 1.15 for the low-mass binary to 1.21 for the high-mass binary. The latter broadens both transition intervals (gains of 0.93 and 0.91), giving a geometric-mean gain of 0.99, compared with 1.04 for the low-mass binary. A 5% moment-of-inertia measurement gives a thickness gain of 1.09. The 10% forecast leaves the widths nearly unchanged.

Across the fourteen forecasts, the median gain is 1.13 for thickness and lies between 0.98 and 1.02 for every other quantity. Of the fifteen gains that exceed 1.05 before rounding, thirteen are for thickness and two for transition density. All fifteen have central 95% bootstrap intervals for GXG_X above unity. None concerns PtP_{\mathrm{t}}, McrustM_{\mathrm{crust}}, or Icrust/II_{\mathrm{crust}}/I. The limited effect on the marginals is also reflected in the Kullback–Leibler divergence, which compares the one-dimensional marginal posterior distributions. Its largest value among the 70 forecast comparisons is 0.16 bits.

Correlations between global observables and crust properties can help interpret these results. In the present-data posterior, we find that the strongest Spearman rank-correlation coefficient ρ\rho between a global observable and crust thickness reaches ∣ρ∣=0.69\lvert\rho\rvert= 0.69, for Λ1.6\Lambda_{1.6} and ΔRcrust(1.4)\Delta R_{\mathrm{crust}}(1.4). The corresponding maximum absolute correlations with ntn_{\mathrm{t}}, PtP_{\mathrm{t}}, McrustM_{\mathrm{crust}}, and Icrust/II_{\mathrm{crust}}/I are 0.40, 0.42, 0.37, and 0.34, respectively. Among the nuclear parameters, LsymL_{\mathrm{sym}} correlates most strongly with ntn_{\mathrm{t}} and PtP_{\mathrm{t}} (ρ=−0.75\rho= -0.75 and −0.59-0.59), while QsymQ_{\mathrm{sym}} has the strongest correlation with the three stellar-crust quantities (ρ=0.65\rho= 0.65–0.67). These correlations are consistent with the stronger forecast constraints on crust thickness.

Table 1 compares the best forecast for each quantity with the changes caused by the nuclear model. Only thickness reaches a gain above 1.20. The best gain for ntn_{\mathrm{t}} is 1.11, and the others do not exceed 1.04. Relative to these forecast widths, removing χ\chiEFT gives intervals 2.58–3.01 times as wide, while the difference between surface models is 0.62–0.84 widths. The forecast improvement is therefore smaller than the sensitivity to the low-density nuclear description.

TargetBest future caseGXG_XDKLD_{\mathrm{KL}} [bits]Δp/W68present+future\Delta_p/W_{68}^{\mathrm{present+future}}W68no χEFT/W68present+futureW_{68}^{\mathrm{no\ \chi EFT}}/W_{68}^{\mathrm{present+future}}
ntn_tmulti-star combined1.110.0800.782.58
PtP_tfive R(M)R(M), 2%1.040.0220.763.01
ΔRcrust(1.4)\Delta R_{\mathrm{crust}}(1.4)Λ1.4+1.4\Lambda_{1.4+1.4}, 5%1.270.1240.622.65
Mcrust(1.4)M_{\mathrm{crust}}(1.4)five R(M)R(M), 2%1.030.0640.832.98
(Icrust/I)(1.4)(I_{\mathrm{crust}}/I)(1.4)five R(M)R(M), 2%1.010.0500.842.99

Table 1. Forecast with the largest width contraction for each crust quantity.

Combined measurements and residual model dependence

To test whether combining measurements improves constraints on the other crust properties, we consider five mass–radius measurements, a single mass–radius measurement combined with tidal deformability and moment of inertia (the “realistic combined” case), and all five mass–radius measurements with the same tidal and inertia measurements (the “multi-star combined” case). The masses and uncertainties are specified in Sect. 3.2. Each combination is added to the full present-data constraints.

Comparisons at the same measurement precision show what each combination adds (Table D.4). The 10% tidal-only forecast for a 1.4 + 1.4 M⊙M_\odot binary gives a thickness gain of 1.17. Adding the 2% mass–radius measurement at 1.8 M⊙M_\odot and the 10% moment-of-inertia measurement raises this to 1.21. The width reduction relative to present data thus increases from about 15% to 17%, or an additional narrowing of about 3% relative to the tidal-only forecast. Compared with the radius-only and inertia-only forecasts at these same precisions, the combined thickness interval is about 8% and 18% narrower, respectively. For the other four crust quantities, the realistic combination gives gains of 0.94–1.00 relative to present data and does not produce a comparable improvement.

Combining five mass–radius measurements gives a thickness gain of 1.22, compared with 1.09–1.12 for the individual radius forecasts considered here. Adding the tidal and inertia measurements leaves this gain unchanged at the reported precision. The main additional improvement is in transition density, whose gain rises from 1.05 to 1.11, corresponding to about 5% further narrowing. The transition-pressure, crust-mass, and crustal-inertia intervals instead become about 2–3% wider. Relative to Notes. The gain is GX=W68present/W68present+futureG_X = W_{68}^{\mathrm{present}}/W_{68}^{\mathrm{present+future}}, and DKLD_{\mathrm{KL}} is measured relative to the full present-data posterior. The last two columns compare the forecast width with Δp=∣q50p=3.5(X)−q50p=2.5(X)∣\Delta_p = \lvert q_{50}^{p=3.5}(X) - q_{50}^{p=2.5}(X)\rvert, the median separation between the two surface models, and with the present width obtained without χ\chiEFT.

the realistic combination, the multi-star programme narrows the transition-density interval by about 15%, but the thickness interval by less than 1%. These comparisons show that combining observables does not substantially extend the improvement beyond crust thickness, apart from a modest gain in transition density. They also show that the five-radius programme already captures most of the thickness improvement in the combined forecast. The stronger constraint on thickness is consistent with its close relation to stellar radius and compactness, while the transition pressure, crust mass, and crustal moment of inertia remain more dependent on sub-saturation matter and surface physics.

Figure 5 shows representative posteriors. The reference crust lies within every central 90% interval, but below 69 of the 70 central 68% intervals. The reference values lie at the 7th–12th percentiles of the present posterior and at the 6th–17th percentiles after adding the forecast data. Although selected near the nuclear-parameter medians, the reference EOS has crust properties near the lower tail of the present posterior. The forecast measurements shift these distributions only slightly towards the reference values.

Figure 5. Representative forecasts for the five crust quantities. Selected single radius, tidal, and moment-of-inertia measurements are shown with the five-radius and combined programmes. Thin and thick segments mark the central 90% and 68% intervals, open circles mark the medians, and the dashed red line marks the crust properties of the reference EOS used to generate all mock measurements.

Forecasts without the χ\chiEFT constraint

Removing χ\chiEFT broadens the allowed low-density EOS range. Four of the forecasts are repeated with the same measurements but without χ\chiEFT to test whether stellar data can recover the lost precision. The χ\chiEFT constraint extends to 0.20 fm−3^{-3}, covering both the crust and part of the uniform-matter regime. Removing it therefore tests the combined contribution of low-density nuclear theory, rather than isolating information about the crust alone. The comparison uses the width ratios defined in Eq. (5).

None of the 20 central 68% intervals narrows relative to the present-data calculation without χEFT\chi_{\mathrm{EFT}}: GXobs=0.79G^{\mathrm{obs}}_X = 0.79–0.960.96 (Figure 4). The full distributions remain close to those obtained before adding the forecast measurements, with DKL≤0.103D_{\mathrm{KL}} \le0.103 bits. Using central 90% intervals also gives no narrowing. The wider intervals should not be interpreted as observations reducing the available information. They describe the response of this interval measure to measurements centred on the chosen reference EOS. We have not established whether the same behaviour persists for other reference EOSs.

Forecastntn_tPtP_tΔRcrust\Delta R_{\mathrm{crust}}McrustM_{\mathrm{crust}}Icrust/II_{\mathrm{crust}}/I
Five R(M)R(M), 2%2.813.343.153.463.52
Λ~1.4+1.4\widetilde{\Lambda}_{1.4+1.4}, 5%2.673.062.763.133.22
Realistic combined2.793.092.723.173.19
Multi-star combined3.073.353.213.393.44

Table 2. Crust uncertainty in forecasts without and with χEFT\chi_{\mathrm{EFT}}.

Notes. Entries give the ratio of the central 68% width without χEFT\chi_{\mathrm{EFT}} to that with χEFT\chi_{\mathrm{EFT}}, RcrustmatchedR^{\mathrm{matched}}_{\mathrm{crust}} (Eq. (5)). Both calculations include the same future measurements and all other present constraints. The stellar crust quantities are evaluated at 1.4 M⊙1.4\,M_\odot.

Without χEFT\chi_{\mathrm{EFT}}, the forecast widths remain 2.17–3.50 times as wide as the current widths obtained with χEFT\chi_{\mathrm{EFT}} and 2.67–3.52 times as wide as the corresponding forecasts retaining $\chi_{\mathrm{EFT}} ({xref}Table 2`). Even a 5% tidal measurement leaves the thickness interval 2.17 times as wide as the current result. All central 90% intervals contain the reference crust.

Distinguishing crust doppelgängers

The doppelgänger test asks whether EOSs with similar global stellar properties can have distinguishable crusts.

We compare EOSs from the full present-data posterior that support all required global observables. We quantify their differences in global observables and crust properties with two dimensionless distances,

dsel2(A,B)=(OA−OB)TCobs−1(OA−OB),(7)d_{\rm sel}^{2}(A,B) = (O_A-O_B)^{T}C_{\rm obs}^{-1}(O_A-O_B), \tag*{(7)}
dcrust2(A,B)=∑X∈ycrust[XA−XB]2sX2,(8)d_{\rm crust}^{2}(A,B) = \sum_{X\in y_{\rm crust}}\frac{[X_A-X_B]^2}{s_X^2}, \tag*{(8)}

Here OAO_A and OBO_B contain the radii, tidal deformabilities, and maximum masses of the two EOSs. The covariance matrix CobsC_{\rm obs} accounts for their spread and correlations in the present posterior, and TT denotes transpose. The quantities XAX_A and XBX_B are their values of crust property XX, and sXs_X is its posterior spread, defined in Appendix C. Small dseld_{\rm sel} and large dcrustd_{\rm crust} identify EOSs with similar global observables but different crusts. We select the pair with the largest dcrustd_{\rm crust} subject to dsel≤0.5d_{\rm sel}\leq0.5, obtaining dsel=0.326d_{\rm sel}=0.326 and dcrust=9.89d_{\rm crust}=9.89. Reducing the allowed global differences to dsel≤0.25d_{\rm sel}\leq0.25 still permits a crust distance of 7.24. Figure C.1 shows how the maximum crust separation varies with this limit.

At the masses used in the forecasts, the global observables differ by only

RA(1.2)−RB(1.2)=5.96×10−3 km,R_A(1.2)-R_B(1.2)=5.96\times10^{-3}\,\mathrm{km},
ΛA(1.2)−ΛB(1.2)=2.89,\Lambda_A(1.2)-\Lambda_B(1.2)=2.89,
IA(1.338)−IB(1.338)=−2.44×1042 g cm2.(9)I_A(1.338)-I_B(1.338)=-2.44\times10^{42}\,\mathrm{g\,cm^2}. \tag*{(9)}

For an equal-mass binary, Λ~=Λ\widetilde{\Lambda}=\Lambda at the component mass. The crust properties nevertheless differ substantially (Table 3). Their separation ranges from 1.8 to 2.5 times the present central 68% width across the five crust quantities.

Crust quantityEOS AAEOS BB∣B−A∣/W68present\lvert B-A\rvert/W_{\rm68}^{\rm present}
ntn_t [fm−3^{-3}]0.072070.099561.80
PtP_t [MeV fm−3^{-3}]0.26010.62852.51
ΔRcrust(1.4)\Delta R_{\rm crust}(1.4) [km]0.88981.11971.84
Mcrust(1.4)M_{\rm crust}(1.4) [M⊙M_\odot]0.014670.033762.37
(Icrust/I)(1.4)(I_{\rm crust}/I)(1.4)0.021570.048322.41

Table 3. Crust properties of the selected doppelgänger EOSs.

The measurements used to test this pair are listed in Table C.1. We express their separation through the dimensionless distance D\mathcal{D}, which measures the difference between the two EOS predictions relative to the assumed measurement uncertainties, accounting for correlations between observables:

Notes. The stellar crust quantities are evaluated at 1.4 M⊙1.4\,M_\odot. The final column gives the absolute difference between EOS BB and AA in units of the full present-data central 68% width, W68presentW_{\rm68}^{\rm present}, for each quantity.

Dk2(A,B)=(μA−μB)TΣk−1(μA−μB).(10)\mathcal{D}_k^2(A,B)=(\mu_A-\mu_B)^T\Sigma_k^{-1}(\mu_A-\mu_B). \tag*{(10)}

Here μA\mu_A and μB\mu_B are the predicted measurement vectors, and Σk\Sigma_k is the assumed covariance for measurement type kk. At the adopted uncertainties, these differences correspond to distances DR=0.025\mathcal{D}_R=0.025, DΛ~=0.030\mathcal{D}_{\widetilde{\Lambda}}=0.030, and DI=0.018\mathcal{D}_I=0.018. Combining independent measurements gives Dcombined=0.043\mathcal{D}_{\rm combined}=0.043. Reaching D=1\mathcal{D}=1 would require reducing the mass–radius, tidal, and inertia uncertainties by factors of 40, 33, and 57, respectively, or by a factor of 23 when all three are combined. Table 4 also gives the factors for D=3\mathcal{D}=3. These distances measure how well the adopted observations separate two fixed EOSs. They do not give a detector significance.

ClassAdopted fractional uncertaintyD0\mathcal{D}_0F1F_1F3F_3
Mass–radius2% in both MM and RR0.024840.3120.8
Tidal10% in Λ~\tilde{\Lambda}0.029933.4100.3
Inertia10% in II0.017656.8170.4
Combined2%, 10%, 10%0.042723.470.3

Table 4. Precision needed to separate the selected crust doppelgängers.

For mass and radius, a factor of 40 corresponds to reducing the uncertainty from 2% to about 0.05%, well below the 5% assumed in a representative STROBE-X forecast [65]. Extending the same fractional tidal precision to 2.0 M⊙M_\odot raises the separation only to $\mathcal{D}=0.43. This is optimistic because tidal uncertainties are expected to increase at higher mass.

The small measurement distances already indicate that the selected EOSs cannot be distinguished at the adopted precision. The posterior recovers test whether this ambiguity persists when the remaining EOS parameters are allowed to vary.

Notes. D0\mathcal{D}_0 is their distance at the adopted uncertainty, and FN=N/D0F_N = N/\mathcal{D}_0 is the factor by which every quoted standard deviation must decrease to reach a fixed-hypothesis distance NN. These detector-independent factors are not detection significances.

To compare the inferred crusts, we project their properties onto the scaled line joining AA and BB,

α(y)=∑X(yX−aX)(bX−aX)/wX2∑X(bX−aX)2/wX2,(11)\alpha(y) = \frac{\sum_X (y_X-a_X)(b_X-a_X)/w_X^2}{\sum_X (b_X-a_X)^2/w_X^2}, \tag*{(11)}

where aXa_X, bXb_X, and yXy_X are the values of crust quantity XX for EOSs AA, BB, and an inferred EOS, respectively. The vector yy contains the five crust quantities, and wX=W68present(X)w_X = W^{\mathrm{present}}_{68}(X) puts them on comparable scales. Thus αA=0\alpha_A = 0 and αB=1\alpha_B = 1. The two recovered distributions shown in Figure 6 overlap by 0.954–0.975, with Jensen–Shannon divergences at most approximately 0.0035 bits (Table D.5). The probability assigned to crusts closer to the injected EOS ranges from 0.433 to 0.588. It changes by at most 0.053 from the present-data value. The recovered medians remain at 0.447≤q50(α)≤0.4860.447 \le q_{50}(\alpha) \le0.486, and neither endpoint lies within

Figure 6. Construction and recovery of the crust doppelgängers. The upper left panel shows the separation between the two EOSs in units of the full present-data 68% widths. The upper right panel gives their distance at the adopted precision for each measurement. The lower panels compare the present posterior with the two matched recoveries along the crust coordinate α\alpha, scaled so that EOSs A and B lie at zero and one. Dotted blue and red lines mark these two EOSs.

the central 90% interval: the recovered distributions concentrate at intermediate values of α\alpha, rather than near either injected endpoint. Combining the radius measurement at 1.2 M⊙1.2\,M_\odot, the tidal measurement for the 1.2+1.2 M⊙1.2+1.2\,M_\odot binary, and the inertia measurement at 1.338 M⊙1.338\,M_\odot does not distinguish the injected crusts.

The pair demonstrates that similar global observables can coexist with substantially different crusts within the adopted model. It does not determine how frequently such pairs occur. Measurements of radius at percent-level precision, low-mass tidal deformability at 10% precision, and total moment of inertia at 10% precision can remain compatible with factor-of-two differences in crust mass and crustal inertia. Distinguishing these crusts would require stronger constraints on neutron-rich matter or measurements that probe the crust more directly.

The crust can affect global stellar properties without being uniquely constrained by their measurement. Fortin et al. [26] showed that crust–core matching affects stellar radii and crust thicknesses, while Gamba et al. [27] found that the assumed crust changes radii inferred from GW170817 despite its weak effect on the inferred masses and tidal deformabilities. Bayesian crust calculations also demonstrate the constraining role of neutron-skin data and microscopic neutron-matter calculations (Newton et al. [53]). The tidal degeneracies identified by Raithel & Most [63] further illustrate that different EOSs can share similar global observables. Our results extend this picture by showing that, within a unified model, even combined radius, tidal-deformability, and moment-of-inertia measurements at the assumed precision can leave substantial differences in crust mass and crustal moment of inertia unresolved. Complementary probes may provide information on the crust’s mechanical, superfluid, and thermal properties. Resonant excitation of the crust–core interface mode (ii-mode) during inspiral may trigger a shattering flare [77, 51], while magnetar quasi-periodic oscillations (QPOs), pulsar glitches, and crust cooling probe other aspects of crust physics [39, 74, 72, 75, 13, 7]. Their interpretation requires additional modelling of composition, magnetic fields, superfluidity, entrainment, and mode identification.

Conclusions

A precise determination of global neutron-star properties does not necessarily imply a precise determination of the crust. Within our unified model, present crust constraints are driven mainly by χ\chiEFT, with laboratory data providing additional information and current NICER and GW170817 observations having a weaker effect. This pattern persists across the stellar masses considered. The inferred precision depends on the adopted χ\chiEFT bands, while the treatment of neutron-rich surface energy changes the inferred crust properties without appreciably changing the credible widths at fixed surface prescription.

Future measurements of radius, tidal deformability, and total moment of inertia mainly improve constraints on crust thickness. Combining these measurements gives modest additional gains at matched precision, with some improvement in transition density but little in transition pressure, crust mass, or crustal moment of inertia. In the forecasts considered, stellar measurements do not recover the precision lost when χ\chiEFT is removed.

These findings are conditional on the chosen reference EOS and the adopted nuclear model. The doppelgänger recoveries further show that substantially different crusts can remain compatible with nearly identical global observables under the tested combination of low-mass radius, tidal, and moment-of-inertia measurements.

The conclusions apply to smooth nucleonic EOSs with a fixed outer crust and a spherical CLDM. We expect the main qualitative conclusion, that global stellar measurements leave substantial uncertainty in crust properties, to persist when nuclear pasta or alternative nucleonic models are considered. Progress therefore requires better constraints on low-density neutron-rich matter and its surface energy, together with observations that probe the crust more directly. Resonant shattering flares, magnetar oscillations, pulsar glitches, and crust cooling offer complementary information beyond global stellar measurements.

Unified-model equations

Fourth-order nuclear metamodel

At zero temperature, we describe uniform nucleonic matter by the baryon density n=nn+npn = n_n + n_p, asymmetry δ=(nn−np)/n=1−2Yp\delta= (n_n - n_p)/n = 1 - 2Y_p, and x=(n−nsat)/(3nsat)x = (n - n_{\mathrm{sat}})/(3n_{\mathrm{sat}}). Here nnn_n and npn_p are neutron and proton number densities, Yp=np/nY_p = n_p/n is the proton fraction, and xx is the scaled displacement from saturation. The energy per baryon is

e(n,δ)=t∗(n,δ)+∑j=04xjj!(vjis+vjivδ2)uj(4)(x).(12)e(n,\delta) = t^*(n,\delta) + \sum_{j=0}^{4} \frac{x^j}{j!}\left(v_j^{\mathrm{is}} + v_j^{\mathrm{iv}}\delta^2\right)u_j^{(4)}(x). \tag*{(12)}

Here j=0,…,4j = 0,\ldots,4 labels the order in the density expansion, and the superscript (4) denotes truncation at fourth order. The term t∗t^* is the kinetic contribution, including the effective-mass dependence, and vjisv_j^{\mathrm{is}} and vjivv_j^{\mathrm{iv}} are the isoscalar and isovector interaction coefficients. The empirical parameters defined in Sect. 2 fix them algebraically. The low-density correction factor is

uj(4)(x)=1−(−3x)5−jexp⁡[−b(1+3x)].(13)u_j^{(4)}(x) = 1 - (-3x)^{5-j}\exp[-b(1+3x)]. \tag*{(13)}

This factor ensures that the energy per baryon vanishes at zero density without changing its value or its first four density derivatives at saturation. It follows the low-density correction introduced by Margueron et al. [46]. Their benchmarks additionally adjust the high-order coefficients to reproduce the density dependence of reference EOSs, demonstrating the flexibility of the fourth-order expansion. These benchmarks do not assign an uncertainty to terms beyond fourth order. Because we do not repeat the inference at another order or with a different low-density correction, the reported credible intervals do not include this model dependence. Cold, fully equilibrated core matter contains neutrons, protons, electrons, and muons in beta equilibrium and satisfies charge neutrality. Accepted EOSs must have positive symmetry energy, mechanical stability, thermodynamic consistency, and subluminal sound speed throughout the density range used by the stellar calculation.

Compressible liquid-drop crust

In a spherical Wigner–Seitz cell, the CLDM energy density, minimized over composition and geometry, is

Ecell=uEbulk(ni,δi)+(1−u)En(ng)+Ee+Esurf+Ecurv+ECoul.(14)\mathcal{E}_{\mathrm{cell}} = u\mathcal{E}_{\mathrm{bulk}}(n_i,\delta_i) + (1-u)\mathcal{E}_n(n_g) + \mathcal{E}_e + \mathcal{E}_{\mathrm{surf}} + \mathcal{E}_{\mathrm{curv}} + \mathcal{E}_{\mathrm{Coul}}. \tag*{(14)}

where uu is the cluster volume fraction, nin_i and δi\delta_i are the baryon density and isospin asymmetry of the cluster interior, respectively, and ngn_g is the dripped-neutron density. The first three terms are the bulk cluster, neutron-gas, and electron energy densities. The remaining terms are the surface, curvature, and Coulomb costs of a finite cluster. We vary the composition and geometry while conserving baryon number and charge. The absence of dripped neutrons distinguishes the outer crust, for which we use the fixed EOS described in Sect. 2.

Writing the cluster proton fraction as Yp,i=(1−δi)/2Y_{p,i} = (1-\delta_i)/2, we use the surface tension

σs(Yp,i)=σ02p+1+bsYp,i−p+bs+(1−Yp,i)−p.(15)\sigma_s(Y_{p,i}) = \sigma_0\frac{2^{p+1}+b_s}{Y_{p,i}^{-p}+b_s+(1-Y_{p,i})^{-p}}. \tag*{(15)}

where σ0\sigma_0 is the surface tension of symmetric matter and bsb_s controls its dependence on proton fraction. The exponent pp sets the neutron-rich extrapolation. The curvature coefficient is σc=σs(σ0,c/σ0)αc(β−Yp,i)\sigma_c = \sigma_s(\sigma_{0,c}/\sigma_0)\alpha_c(\beta-Y_{p,i}), where σ0,c\sigma_{0,c} sets its scale, β\beta controls its proton-fraction dependence, and αc=5.5\alpha_c = 5.5 is fixed (Newton et al. [52]). For every sampled bulk EOS, we refit the coefficients {σ0,bs,σ0,c,β}\{\sigma_0,b_s,\sigma_{0,c},\beta\} by least squares to masses from the 2020 Atomic Mass Evaluation (AME2020) (Wang et al. [79]). The exponent pp is fixed within each calculation. We use p=3p = 3 as the reference and repeat the calculation at p=2.5p = 2.5 and 3.53.5 to measure the sensitivity to the surface energy of very neutron-rich nuclei. The crust–core transition occurs where uniform beta-equilibrated matter becomes energetically favoured over the optimized nuclear phase.

Priors and likelihoods

Prior and present-data likelihood

For data DD and model M\mathcal{M}, Bayes’ theorem gives

p(θ∣D,M)=L(D∣θ,M)π(θ∣M)Z(D∣M),Z=∫Lπ dθ.(16)p(\theta|D,\mathcal{M}) = \frac{\mathcal{L}(D|\theta,\mathcal{M})\pi(\theta|\mathcal{M})}{\mathcal{Z}(D|\mathcal{M})}, \qquad\mathcal{Z} = \int\mathcal{L}\pi\,\mathrm{d}\theta. \tag*{(16)}

Here p(θ∣D,M)p(\theta|D,\mathcal{M}) is the posterior density, π\pi is the prior density, L\mathcal{L} is the likelihood, and the evidence Z\mathcal{Z} normalizes the posterior. We use Z\mathcal{Z} only for normalization, not to compare cases that contain different data. Table B.1 lists the independent uniform priors on the metamodel parameters.

ParameterPriorUnitParameterPriorUnit
nsatn_{\mathrm{sat}}[0.150, 0.170]fm−3^{-3}EsatE_{\mathrm{sat}}[-17, -15]MeV
KsatK_{\mathrm{sat}}[190, 320]MeVQsatQ_{\mathrm{sat}}[-1000, 1000]MeV
ZsatZ_{\mathrm{sat}}[-3000, 3000]MeVJsymJ_{\mathrm{sym}}[25, 40]MeV
LsymL_{\mathrm{sym}}[10, 100]MeVKsymK_{\mathrm{sym}}[-400, 400]MeV
QsymQ_{\mathrm{sym}}[-2000, 2000]MeVZsymZ_{\mathrm{sym}}[-5000, 5000]MeV
m∗/mm^*/m[0.60, 0.80]—Δm∗\Delta m^*[0, 0.20]—
bb[1, 10]—

Table B.1. Continuous uniform priors.

Notes. Energy coefficients are in MeV. The effective-mass parameters are dimensionless and defined in Sect. 2.

The physical term in Eq. (2) requires a valid unified EOS that satisfies the conditions in Appendix § 7. We treat the observed maximum mass as a lower bound. EOSs with Mmax⁡M_{\max} well above 2 M⊙2\,M_\odot receive the same weight, while a one-sided Gaussian turn-on of width 0.02 M⊙0.02\,M_\odot smoothly suppresses models below it. threshold is consistent with the observed two-solar-mass pulsars [5, 18]. Table B.2 summarizes all terms in the present-data likelihood. For the Gaussian terms in Lsat\mathcal{L}_{\mathrm{sat}} and Llab\mathcal{L}_{\mathrm{lab}}, a quoted value μ±σ\mu\pm\sigma contributes a normal density N[Xθ(n);μ,σ]\mathcal{N}[X_{\theta}(n);\mu,\sigma], where μ\mu is the measured central value, σ\sigma its standard deviation, and Xθ(n)X_{\theta}(n)

the corresponding prediction of the sampled EOS at density nn. The adopted saturation constraints and their supporting sources are listed in Table B.2. The finite-density set is the subset of the experimental compilation of Tsang et al. [76] used in our calculation. The laboratory constraints are treated as independent.

LikelihoodPhysical inputnn [fm−3^{-3}]Adopted constraint or likelihoodSource
Lphys\mathcal{L}_{\mathrm{phys}}Physical admissibilityFull EOSThermodynamic consistency, stability, positive symmetry energy, and causalityAppendix A
Lsat\mathcal{L}_{\mathrm{sat}}Saturation density nsatn_{\mathrm{sat}}—0.155±0.0050.155 \pm0.005 fm−3^{-3}[8, 46]
Saturation energy EsatE_{\mathrm{sat}}—−16.0±0.2-16.0 \pm0.2 MeV[12, 68]
Incompressibility KsatK_{\mathrm{sat}}—230±40230 \pm40 MeV[40]
Llab\mathcal{L}_{\mathrm{lab}}Dipole polarizability: S(n)S(n)0.0500.05015.9±1.015.9 \pm1.0[81]
Isobaric analogue states and isovector skins: S(n)S(n)0.1060.10625.5±1.125.5 \pm1.1[20]
Isospin diffusion: S(n)S(n)0.0350.03510.3±1.010.3 \pm1.0[78]
Neutron/proton yields: S(n)S(n)0.0690.06916.8±1.216.8 \pm1.2Morfouace et al. (2019)
Pion spectral ratio: S(n)S(n)0.2320.23252.0±13.052.0 \pm13.0[24]
Pion spectral ratio: Psym(n)P_{\mathrm{sym}}(n)0.2320.23210.9±8.710.9 \pm8.7[24]
Neutron/proton elliptic flow: Psym(n)P_{\mathrm{sym}}(n)0.2400.24012.1±8.412.1 \pm8.4[17, 70, 71]
Collective flow: PSNM(n)P_{\mathrm{SNM}}(n)0.3200.32010.1±3.010.1 \pm3.0[19]
LχEFT\mathcal{L}_{\chi\mathrm{EFT}}Symmetric and pure-neutron matter energies0.020.02–0.200.20Uniform acceptance inside the energy bands at the evaluated densities, zero outside. Reference widening f=0.05f=0.05 (Eq. (B.3))[22, 10 b]
LMmax⁡\mathcal{L}_{M_{\max}}Maximum stellar mass—One-sided Gaussian suppression below 2 M⊙2\,M_{\odot}, with width 0.02 M⊙0.02\,M_{\odot}[5, 18]
LNICER\mathcal{L}_{\mathrm{NICER}}PSR J0030+0451: M,RM,R—M=1.34−0.16+0.15 M⊙M=1.34^{+0.15}_{-0.16}\,M_{\odot}, R=12.71−1.19+1.14R=12.71^{+1.14}_{-1.19} km[66]
PSR J0437−4715: M,RM,R—M=1.418±0.037 M⊙M=1.418 \pm0.037\,M_{\odot}, R=11.36−0.63+0.95R=11.36^{+0.95}_{-0.63} km[15]
PSR J0614−3329: M,RM,R—M=1.44−0.07+0.06 M⊙M=1.44^{+0.06}_{-0.07}\,M_{\odot}, R=10.29−0.86+1.01R=10.29^{+1.01}_{-0.86} km[49]
PSR J0740+6620: M,RM,R—M=2.072−0.066+0.067 M⊙M=2.072^{+0.067}_{-0.066}\,M_{\odot}, R=12.39−0.98+1.30R=12.39^{+1.30}_{-0.98} km[67]
LGW170817\mathcal{L}_{\mathrm{GW170817}}Binary tidal deformability Λ~\tilde{\Lambda} and mass ratio qq—Public EOS-insensitive posterior in (q,Λ)(q,\Lambda), marginalized over 0.7≤q≤10.7 \le q \le1 at Mc=1.186 M⊙\mathcal{M}_{c}=1.186\,M_{\odot}[1]

Table B.2. Nuclear and astrophysical inputs to the present-data likelihood.

Notes. S(n)S(n) is the symmetry energy, L(n)L(n) is its local density slope, Psym(n)P_{\mathrm{sym}}(n) is the symmetry pressure, and PSNM(n)P_{\mathrm{SNM}}(n) is the pressure of symmetric nuclear matter. Energies are in MeV and pressures in MeV fm−3^{-3}. For the nuclear Gaussian terms, quoted uncertainties are one standard deviation. The χEFT\chi\mathrm{EFT} widening is applied to the absolute boundary energies. Sensitivity variants use f=0f=0, 0.100.10, 0.200.20, or omit this term. NICER values summarize the published marginal 68%68\% credible intervals for mass and equatorial radius. Each NICER source density is integrated along the predicted mass–radius curve using Eq. (B.4), and the four source terms are multiplied. The chirp mass Mc\mathcal{M}_{c}, mass ratio qq, and tidal parameter Λ~\tilde{\Lambda} are defined below.

We compare symmetric and pure-neutron matter with the χEFT\chi\mathrm{EFT} envelope at densities from 0.020.02 to 0.200.20 fm−3^{-3}.

ln⁡LχEFT={0,elow(nj)≤eθ(nj)≤ehigh(nj)∀j,−∞,otherwise.(17)\ln\mathcal{L}_{\chi\mathrm{EFT}} = \begin{cases} 0, & e_{\mathrm{low}}(n_j) \le e_{\theta}(n_j) \le e_{\mathrm{high}}(n_j)\quad\forall j,\\ -\infty, & \text{otherwise.} \end{cases} \tag*{(17)}

Here njn_j denotes an evaluated density, with jj indexing these densities. The quantity eθe_{\theta} is the model energy per baryon, and elowe_{\mathrm{low}} and ehighe_{\mathrm{high}} are the lower and upper acceptance limits. The inequalities must hold for both symmetric and pure-neutron matter. We use the bands published with the NSEoS implementation (Carreau et al. [10]). They draw on the chiral two- and three-nucleon calculations of Drischler et al. [22]. Every EOS inside the envelope has the same likelihood. In the reference calculation we widen both edges outwards by a fraction f=0.05f=0.05 of their absolute energies:

elow′=elow−f∣elow∣,ehigh′=ehigh+f∣ehigh∣.(18)e'_{\mathrm{low}} = e_{\mathrm{low}} - f\lvert e_{\mathrm{low}}\rvert,\qquad e'_{\mathrm{high}} = e_{\mathrm{high}} + f\lvert e_{\mathrm{high}}\rvert. \tag*{(18)}

The primed limits are used in Eq. (B.2). Thus the percentage refers to the boundary energies, not to the original band width. The sensitivity calculations use the original χ\chiEFT bands (f=0f = 0), widenings of f=0.10f = 0.10 and 0.200.20, or omit the χ\chiEFT constraint entirely.

For each NICER source ss, we evaluate the reconstructed mass–radius density along the EOS curve,

Ls(θ)∝∫dM ps[M,Rθ(M)∣Ds].(19)\mathcal{L}_s(\theta) \propto\int\mathrm{d}M\,p_s[M,R_\theta(M)\mid D_s]. \tag*{(19)}

Here Rθ(M)R_\theta(M) is the predicted radius, DsD_s denotes the observations included in the published source analysis, and psp_s is its reconstructed joint mass–radius posterior density. This prescription uses a flat reference prior in mass and radius. Any radio timing information included in DsD_s is retained in psp_s. We include J0030+0451 (Riley et al. [66]), J0437–4715 (Choudhury et al. [15]), J0614–3329 (Mauviard et al. [49]), and J0740+6620 (Riley et al. [67]).

For GW170817, we use the public EOS-insensitive posterior in the mass ratio q=m2/m1≤1q = m_2/m_1 \le1 and the binary tidal parameter Λ~\tilde{\Lambda} (Abbott et al. [1]). Here m1m_1 and m2m_2 are the source-frame component masses, and

Λ~=1613(m1+12m2)m14Λ1+(m2+12m1)m24Λ2(m1+m2)5.(20)\tilde{\Lambda} = \frac{16}{13}\frac{(m_1+12m_2)m_1^4\Lambda_1+(m_2+12m_1)m_2^4\Lambda_2}{(m_1+m_2)^5}. \tag*{(20)}

Here Λ1=Λ(m1)\Lambda_1 = \Lambda(m_1) and Λ2=Λ(m2)\Lambda_2 = \Lambda(m_2) are evaluated from the sampled EOS. We marginalize over q≥0.7q \ge0.7 at fixed source-frame chirp mass Mc=(m1m2)3/5/(m1+m2)1/5=1.186 M⊙\mathcal{M}_c = (m_1m_2)^{3/5}/(m_1+m_2)^{1/5} = 1.186\,M_\odot. This retains the binary-mass uncertainty while testing the complete tidal curve rather than one value of Λ\Lambda.

Forecast likelihoods and reference EOS

We choose the reference EOS TREFT_{\rm REF} from the full present-data posterior, near the nuclear-parameter medians. We require the reference EOS to support the full forecast mass range, with a small margin above the highest target mass.

For a mass–radius observation centred on μMR=(M⋆,RTREF(M⋆))\boldsymbol{\mu}_{MR} = (M_\star,R_{T_{\rm REF}}(M_\star)), where M⋆M_\star is the injected stellar mass and RTREFR_{T_{\rm REF}} is the reference EOS radius, the model-curve likelihood is

LMR(θ)=∫dM N2([MRθ(M)];μMR,ΣMR).(21)\mathcal{L}_{MR}(\theta) = \int\mathrm{d}M\,\mathcal{N}_2\left(\begin{bmatrix} M \\ R_\theta(M) \end{bmatrix};\boldsymbol{\mu}_{MR},\boldsymbol{\Sigma}_{MR}\right). \tag*{(21)}

where N2\mathcal{N}_2 is the bivariate normal density with centre μMR\boldsymbol{\mu}_{MR} and covariance ΣMR\boldsymbol{\Sigma}_{MR}. The integral follows the predicted Rθ(M)R_\theta(M) curve and therefore marginalizes over the measured mass rather than evaluating the EOS at one fixed mass. We use independent 2% standard deviations in MM and RR. The tidal forecasts use a reference equal-mass binary with component mass m⋆m_\star and a 5%, 10%, or 20% Gaussian uncertainty on Λ~\tilde{\Lambda}. We marginalize over a Gaussian mass-ratio distribution peaking at q=1q = 1, with Gaussian width 0.03, truncated to 0.85≤q≤10.85 \le q \le1, at fixed injected chirp mass. The moment-of-inertia likelihood is

LI(θ)=N[Iθ(1.338 M⊙);ITREF,fIITREF].(22)\mathcal{L}_I(\theta) = \mathcal{N}[I_\theta(1.338\,M_\odot);I_{T_{\rm REF}},f_I I_{T_{\rm REF}}]. \tag*{(22)}

where IθI_\theta is the predicted moment of inertia, ITREF=ITREF(1.338 M⊙)I_{T_{\rm REF}} = I_{T_{\rm REF}}(1.338\,M_\odot) is its injected value, and fI=0.05f_I = 0.05 or 0.10 is the fractional standard deviation. Writing DpresentD_{\rm present} and DfutureD_{\rm future} for the present and forecast data, with Lfuture\mathcal{L}_{\rm future} denoting the likelihood of the added measurements, each forecast samples

p(θ∣Dpresent,Dfuture)∝LpresentLfutureπ(θ).(23)p(\theta\mid D_{\rm present},D_{\rm future}) \propto\mathcal{L}_{\rm present}\mathcal{L}_{\rm future}\pi(\theta). \tag*{(23)}

The corresponding no-χ\chiEFT posterior is

p¬χEFT(θ∣Dpresent,Dfuture)∝Lpresent¬χEFTLfutureπ(θ),(24)p_{\neg\chi{\rm EFT}}(\theta\mid D_{\rm present},D_{\rm future}) \propto\mathcal{L}_{\rm present}^{\neg\chi{\rm EFT}}\mathcal{L}_{\rm future}\pi(\theta), \tag*{(24)}

where the superscript ¬χ\neg\chiEFT denotes omission of the χ\chiEFT factor from the present likelihood. The reference EOS and future likelihood Lfuture\mathcal{L}_{\rm future} are unchanged. We repeat four forecasts: five 2% radius measurements, a 5% Λ~1.4+1.4\tilde{\Lambda}_{1.4+1.4} measurement, the realistic combined programme, and the multi-star combined programme. Each forecast differs from its χ\chiEFT-conditioned counterpart only through removal of the χ\chiEFT term.

Crust-doppelgänger definition

The observable vector O\boldsymbol{O} in Eq. (7) contains radii at 1.2, 1.4, and 1.8 M⊙M_\odot, tidal deformabilities at 1.2, 1.4, and 1.6 M⊙M_\odot, and the maximum mass MmaxM_{\rm max}. Moment of inertia is reserved for an independent test. The selection distances are defined in Eqs. (7) and (8). The scale sX=1.4826 median⁡∣X−median⁡(X)∣s_X = 1.4826\,\operatorname{median}|X-\operatorname{median}(X)|, evaluated with posterior weights, measures the spread while reducing sensitivity to extreme values. The factor 1.4826 gives the standard deviation for a Gaussian distribution (Rousseeuw & Croux [69]). We maximize dcrustd_{\rm crust} subject to dsel≤0.5d_{\rm sel} \le0.5. Figure C.1 shows the Pareto frontier (the best compromises between similar global observables and different crusts) (Marler & Arora [48]).

The measurement distance is defined in Eq. (10). Independent measurements add in D2\mathcal{D}^2. For a scalar observable OO, the fractional standard deviation needed to reach a distance NN is

freq(N)=∣OA−OB∣NOˉ,Oˉ=OA+OB2.(25)f_{\rm req}(N)=\frac{|O_A-O_B|}{N\bar{O}},\qquad\bar{O}=\frac{O_A+O_B}{2}. \tag*{(25)}

Here OAO_A and OBO_B are the two predictions. Reducing all standard deviations by FN=N/DKF_N=N/D_K gives the same target distance (Table 4). These distances compare fixed EOSs and are not detection significances for a particular instrument.

The measurements in Table C.1 were chosen before the recoveries. For radius and tidal deformability, we selected the forecast with the smallest joint covariance volume of the five scaled crust quantities. The inertia precision follows the reference forecast for J0737–3039A. Both injected EOSs use the same measurement uncertainties.

ClassAdded zero-noise measurement
Mass–radiusM,RM,R at 1.2 M⊙M_\odot with independent 2% errors
TidalEqual-mass 1.2+1.2 M⊙1.2+1.2\,M_\odot binary with 10% uncertainty in Λ~\tilde{\Lambda} and a Gaussian mass-ratio distribution peaking at q=1q = 1, with Gaussian width 0.03, truncated to 0.85≤q≤10.85 \le q \le1
InertiaI(1.338 M⊙)I(1.338\,M_\odot) with 10% uncertainty
Combinedall three measurements above

Table C.1. Forecast measurements for the crust doppelgängers.

Notes. We generate each measurement once from EOS A and once from EOS B. We always combine it with the full present-data constraints.

We compare the recovered crusts using the coordinate α\alpha defined in Eq. (11). For each measurement type, let pA(α)p_A(\alpha) and pB(α)p_B(\alpha) be the normalized posterior densities obtained from data generated by A and B, respectively. We report their posterior-overlap coefficient (Pastore & Calcagni [60])

OAB=∫min⁡[pA(α),pB(α)] dα.(26)O_{AB}=\int\min[p_A(\alpha),p_B(\alpha)]\,\mathrm{d}\alpha. \tag*{(26)}

which runs from zero for disjoint distributions to one for identical distributions. We also use the symmetric Jensen–Shannon divergence DJSD_{\mathrm{JS}} Lin [44].

DJS=12DKL(pA∥pmix)+12DKL(pB∥pmix),D_{\mathrm{JS}} = \frac{1}{2}D_{\mathrm{KL}}(p_A\Vert p_{\mathrm{mix}}) + \frac{1}{2}D_{\mathrm{KL}}(p_B\Vert p_{\mathrm{mix}}),
pmix=pA+pB2,(27)p_{\mathrm{mix}} = \frac{p_A+p_B}{2}, \tag*{(27)}

Here DKL(p∥q)=∫p(α)log⁡2[p(α)/q(α)] dαD_{\mathrm{KL}}(p\Vert q) = \int p(\alpha)\log_{2}[p(\alpha)/q(\alpha)]\,\mathrm{d}\alpha is the Kullback–Leibler divergence of Eq. (4), applied to the normalized densities in α\alpha, and pmixp_{\mathrm{mix}} is their equally weighted mixture. The Jensen–Shannon divergence is symmetric and ranges from zero for identical distributions to one bit for distributions with no overlap. We also report the probability on the side of α=0.5\alpha= 0.5 containing the injected EOS: α<0.5\alpha< 0.5 for A and α>0.5\alpha> 0.5 for B. Recovery requires consistency with the injected EOS and preference over the other member, rather than narrowing alone. Table D.5 lists all eight recoveries.

Figure C.1. Robustness of the crust-doppelgänger selection. Left: joint crust distance dcrustd_{\mathrm{crust}} versus exterior-observable distance dseld_{\mathrm{sel}} along the Pareto frontier. Orange points mark the largest crust separations below the tested limits on dseld_{\mathrm{sel}}, and the star marks the selected pair. Right: number of accepted EOS pairs below each exterior-distance limit, shown for all pairs and for three minimum crust separations. The vertical dashed lines mark dsel=0.5d_{\mathrm{sel}} = 0.5. The annotation in the right panel gives the pair counts at this limit.

Numerical posterior tables

These tables summarize the posterior constraints and forecast results.

Parametermedian68%90%Crust quantitymedian68%90%
nsatn_{\mathrm{sat}} [fm−3^{-3}]0.1601[0.1561, 0.1646][0.1537, 0.1671]ntn_t [fm−3^{-3}]0.0811[0.0742, 0.0895][0.0704, 0.0951]
EsatE_{\mathrm{sat}} [MeV]−15.95[−16.13, −15.76][−16.25, −15.65]PtP_t [MeV fm−3^{-3}]0.396[0.328, 0.475][0.291, 0.528]
KsatK_{\mathrm{sat}} [MeV]231[215, 246][206, 257]ΔRcrust(1.4)\Delta R_{\mathrm{crust}}(1.4) [km]1.048[0.985, 1.110][0.943, 1.151]
QsatQ_{\mathrm{sat}} [MeV]−359[−499, −206][−575, −89]Mcrust(1.4)M_{\mathrm{crust}}(1.4) [M⊙M_{\odot}]0.0243[0.0204, 0.0285][0.0180, 0.0314]
ZsatZ_{\mathrm{sat}} [MeV]674[−230, 1776][−573, 2436](Icrust/I)(1.4)(I_{\mathrm{crust}}/I)(1.4)0.0355[0.0301, 0.0412][0.0267, 0.0452]
JsymJ_{\mathrm{sym}} [MeV]30.69[29.70, 31.79][29.24, 32.43]
LsymL_{\mathrm{sym}} [MeV]47.04[38.37, 54.75][33.39, 58.61]
KsymK_{\mathrm{sym}} [MeV]−85[−137, −32][−171, 6]
QsymQ_{\mathrm{sym}} [MeV]965[613, 1327][397, 1565]
ZsymZ_{\mathrm{sym}} [MeV]−2413[−4042, −406][−4635, 879]
m∗/mm^*/m0.69[0.63, 0.77][0.61, 0.79]
Δm∗\Delta m^*0.11[0.04, 0.17][0.01, 0.19]
bb4.31[1.89, 7.76][1.34, 9.32]

Table D.1. Reference posterior using all present constraints.

Notes. The left block gives the thirteen empirical parameters. The effective-mass parameters and bb are dimensionless. The right block gives the five crust targets, with mass-labelled quantities at 1.4 M⊙1.4\,M_{\odot}. Entries are the median and central 68% and 90% intervals.

Casentn_tPtP_tΔRcrust\Delta R_{\mathrm{crust}}McrustM_{\mathrm{crust}}Icrust/II_{\mathrm{crust}}/Imax⁡∣ΔX~∣/W68present\max|\widetilde{\Delta X}|/W^{\mathrm{present}}_{68}
physical only2.292.393.283.043.191.78
nuclear only0.840.861.111.021.010.15
astrophysical only2.733.542.913.653.780.98
all present1.001.001.001.001.000.00
no χ\chiEFT2.342.912.082.892.960.86
no laboratory1.271.241.111.201.190.32
no NICER0.830.881.050.991.000.13
no GW1708170.980.991.051.031.040.07
χ\chiEFT original0.480.420.780.540.530.25
χ\chiEFT 10%1.281.341.221.371.380.09
χ\chiEFT 20%1.691.961.491.931.970.21
surface p=2.5p = 2.51.020.971.041.031.040.52
surface p=3.5p = 3.50.971.031.011.061.060.32

Table D.2. Sensitivity to present constraints.

Notes. The five central columns give W68case/W68presentW^{\mathrm{case}}_{68}/W^{\mathrm{present}}_{68}. Values above one are broader than the reference posterior. The last column gives the largest absolute normalized median shift, max⁡X∣q50case(X)−q50present(X)∣/W68present(X)\max_X|q^{\mathrm{case}}_{50}(X)-q^{\mathrm{present}}_{50}(X)|/W^{\mathrm{present}}_{68}(X), where ΔX~=q50case(X)−q50present(X)\widetilde{\Delta X}=q^{\mathrm{case}}_{50}(X)-q^{\mathrm{present}}_{50}(X).

M/M⊙M/M_\odotΔRcrust\Delta R_{\mathrm{crust}} [km]McrustM_{\mathrm{crust}} [M⊙M_\odot]Icrust/II_{\mathrm{crust}}/I
1.01.573 [1.492, 1.654]0.0320 [0.0273, 0.0369]0.0641 [0.0555, 0.0733]
1.21.275 [1.204, 1.347]0.0280 [0.0237, 0.0326]0.0472 [0.0403, 0.0544]
1.41.048 [0.985, 1.110]0.0243 [0.0204, 0.0285]0.0355 [0.0301, 0.0412]
1.60.864 [0.809, 0.920]0.0209 [0.0174, 0.0247]0.0269 [0.0226, 0.0314]
1.80.708 [0.659, 0.759]0.0176 [0.0145, 0.0209]0.0203 [0.0169, 0.0240]
2.00.573 [0.525, 0.619]0.0143 [0.0117, 0.0172]0.0151 [0.0124, 0.0180]

Table D.3. Mass dependence of the reference crust posterior.

Notes. Each cell reports the median followed by the central 68% interval [q16,q84q_{16}, q_{84}].

Casentn_tPtP_tΔRcrust\Delta R_{\mathrm{crust}}McrustM_{\mathrm{crust}}Icrust/II_{\mathrm{crust}}/IGgeomG_{\mathrm{geom}}max⁡DKL\max D_{\mathrm{KL}} [bits]
R(1.2)R(1.2), 2%1.051.021.090.960.961.010.027
R(1.4)R(1.4), 2%1.011.001.120.980.981.010.029
R(1.6)R(1.6), 2%0.980.981.120.970.971.000.035
R(1.8)R(1.8), 2%0.970.961.110.990.981.000.065
five R(M)R(M), 2%1.051.041.221.031.011.070.147
Λˉ1.2+1.2\bar{\Lambda}_{1.2+1.2}, 10%1.051.031.151.001.001.040.047
Λˉ1.4+1.4\bar{\Lambda}_{1.4+1.4}, 10%0.970.981.170.990.991.020.055
Λˉ1.6+1.6\bar{\Lambda}_{1.6+1.6}, 10%0.930.911.210.980.970.990.145
Λˉ1.4+1.4\bar{\Lambda}_{1.4+1.4}, 5%0.970.981.271.011.011.040.124
Λˉ1.4+1.4\bar{\Lambda}_{1.4+1.4}, 20%1.020.991.060.990.991.010.017
I(1.338)I(1.338), 10%1.020.980.990.970.960.980.006
I(1.338)I(1.338), 5%1.011.021.091.000.991.020.021
realistic combined0.940.951.211.000.981.010.115
multi-star combined1.111.011.221.010.981.060.162

Table D.4. Effect of future measurements.

Notes. The five crust columns give GX=W68present/W68present+futureG_X = W^{\mathrm{present}}_{68}/W^{\mathrm{present+future}}_{68}. Values above one indicate narrower intervals. GgeomG_{\mathrm{geom}} is their geometric mean, and the final column is the largest marginalized information gain for that measurement.

MeasurementD\mathcal{D}OABO_{AB}DJSD_{\mathrm{JS}} [bits]PA(α<0.5)P_A(\alpha< 0.5)PB(α>0.5)P_B(\alpha> 0.5)max⁡∣ΔP∣\max|\Delta P|
mass–radius0.02480.9540.003510.5240.4330.032
tidal0.02990.9680.002330.5880.4420.053
inertia0.01760.9670.001650.5600.4430.024
combined0.04270.9750.001940.5530.4340.030

Table D.5. Recovery of the selected crust doppellgängers.

Notes. The crust properties of the two EOSs are listed in Table 3. D\mathcal{D} is the fixed-EOS distance at the adopted measurement precision, OABO_{AB} is the overlap of the two recovered distributions in α\alpha, DJSD_{\mathrm{JS}} is their Jensen–Shannon divergence, and PAP_A (PBP_B) is the probability assigned to the correct side of α=0.5\alpha= 0.5 for data generated from AA (BB). The final column is the larger absolute change in this probability relative to the present-data posterior.

References

  1. [1]Abbott, B. P. et al. 2018, Phys. Rev. Lett., 121, 161101
  2. [2]Anderson, P. W. & Itoh, N. 1975, Nature, 256, 25
  3. [3]Andersson, N., Glampedakis, K., Ho, W. C. G., & Espinoza, C. M. 2012, Phys. Rev. Lett., 109, 241103
  4. [4]Annala, E., Gorda, T., Kurkela, A., Nättilä, J., & Vuorinen, A. 2020, Nature Phys., 16, 907
  5. [5]Antoniadis, J. et al. 2013, Science, 340, 1233232
  6. [6]Balliet, L. E., Newton, W. G., Cantu, S., & Budimir, S. 2021, ApJ, 918, 79
  7. [7]Brown, E. F. & Cumming, A. 2009, ApJ, 698, 1020
  8. [8]Buchinger, F., Crawford, J. E., Dutta, A. K., Pearson, J. M., & Tondeur, F. 1994, Phys. Rev. C, 49, 1402
  9. [9]Burrello, S., Gülminelli, F., Antonelli, M., Colonna, M., & Fantina, A. F. 2025, Phys. Rev. C, 112, 035802
  10. [10]Carreau, T., Gulminelli, F., & Margueron, J. 2019a, Eur. Phys. J. A, 55, 188
  11. [12]Chabanat, E., Bonche, P., Haensel, P., Meyer, J., & Schaeffer, R. 1997, Nucl. Phys. A, 627, 710
  12. [13]Chamel, N. 2013, Phys. Rev. Lett., 110, 011101
  13. [14]Chamel, N. & Haensel, P. 2008, Living Rev. Rel., 11, 10
  14. [15]Choudhury, D. et al. 2024, ApJL, 971, L20
  15. [16]Cowan, G., Cranmer, K., Gross, E., & Vitells, O. 2011, Eur. Phys. J. C, 71, 1554
  16. [17]Cozma, M. D. 2018, Eur. Phys. J. A, 54, 40
  17. [18]Cromartie, H. T. et al. 2020, Nat. Astron., 4, 72
  18. [19]Danielewicz, P., Lacey, R., & Lynch, W. G. 2002, Science, 298, 1592
  19. [20]Danielewicz, P., Singh, P., & Lee, J. 2017, Nucl. Phys. A, 958, 147
  20. [21]Dinh Thi, H., Carreau, T., Fantina, A. F., & Gulminelli, F. 2021, A&A, 654, A114
  21. [22]Drischler, C., Hebeler, K., & Schwenk, A. 2016, Phys. Rev. C, 93, 054314
  22. [23]Essick, R., Landry, P., & Holz, D. E. 2020, Phys. Rev. D, 101, 063007
  23. [24]Estee, J. et al. 2021, Phys. Rev. Lett., 126, 162701
  24. [25]Ferreira, M. & Providência, C. 2020, Universe, 6, 220
  25. [26]Fortin, M., Providencia, C., Raduta, A. R., et al. 2016, Phys. Rev. C, 94, 035804
  26. [27]Gamba, R., Read, J. S., & Wade, L. E. 2020, Class. Quant. Grav., 37, 025008
  27. [28]Geppert, U., Küker, M., & Page, D. 2004, A&A, 426, 267
  28. [29]Goriely, S., Chamel, N., Janka, H.-T., & Pearson, J. M. 2011, A&A, 531, A78
  29. [30]Goriely, S., Chamel, N., & Pearson, J. M. 2013, Phys. Rev. C, 88, 024308
  30. [31]Göttling, H., Hoff, L., Hebeler, K., & Schwenk, A. 2026, Phys. Rev. C, 113, 055804
  31. [32]Grams, G., Margueron, J., Somandurdam, R., Chamel, N., & Goriely, S. 2022, J. Phys. Conf. Ser., 2340, 012030
  32. [33]Hebeler, K., Lattimer, J. M., Pethick, C. J., & Schwenk, A. 2013, ApJ, 773, 11
  33. [34]Horowitz, C. J. & Kadauk, K. 2009, Phys. Rev. Lett., 102, 191102
  34. [35]Hotokezaka, K., Kyutoku, K., Sekiguchi, Y., & Shibata, M. 2016, Phys. Rev. D, 93, 064082
  35. [36]Huang, C., Tolos, L., Providencia, C., & Watts, A. L. 2025, MNRAS, 536, 3262
  36. [37]Huth, S., Pang, P. T. H., Tews, I., et al. 2022, Nature, 606, 276
  37. [38]Iacovelli, F., Mancarella, M., Mondal, C., et al. 2023, Phys. Rev. D, 108, 122006
  38. [39]Israel, G. L., Belloni, T., Stella, L., et al. 2005, ApJL, 628, L53
  39. [40]Khan, E., Margueron, J., & Vidaña, I. 2012, Phys. Rev. Lett., 109, 092501
  40. [41]Klausner, P., Antonelli, M., & Gulminelli, F. 2026, Phys. Rev. C, 113, 025808
  41. [42]Kullback, S. & Leibler, R. A. 1951, Ann. Math. Stat., 22, 79
  42. [43]Lim, Y.-H., Holt, J. W., & Stahulak, R. J. 2019, Phys. Rev. C, 100, 035802
  43. [44]Lin, J. 1991, IEEE Transactions on Information Theory, 37, 145
  44. [45]Link, B., Epstein, R. I., & Lattimer, J. M. 1999, Phys. Rev. Lett., 83, 3362
  45. [46]Margueron, J., Hoffmann Casali, R., & Gulminelli, F. 2018a, Phys. Rev. C, 97, 025805
  46. [48]Marler, R. T. & Arora, J. S. 2004, Structural and Multidisciplinary Optimization, 26, 369
  47. [49]Mauviard, L. et al. 2025, ApJ, 995, 60
  48. [51]Neill, D., Preston, R., Newton, W. G., & Tsang, D. 2023, Phys. Rev. Lett., 130, 112701
  49. [52]Newton, W. G., Gearheart, M., & Li, B.-A. 2013, ApJS, 204, 9
  50. [53]Newton, W. G., Preston, R., Balliet, L., & Ross, M. 2022, Phys. Lett. B, 834, 137481
  51. [54]Pacilio, C., Maselli, A., Fasano, M., & Pani, P. 2022, Phys. Rev. Lett., 128, 101101
  52. [55]Page, D. & Reddy, S. 2013, Phys. Rev. Lett., 111, 241102
  53. [56]Pang, P. T. H., Tews, I., Coughlin, M. W., et al. 2021, ApJ, 922, 14
  54. [57]Parmar, V. & Bombaci, I. 2026a, Phys. Rev. D, 113, 083017
  55. [58]Parmar, V. & Bombaci, I. 2026b, Phys. Rev. D, 114, 023009
  56. [59]Parmar, V., Scordino, D., & Bombaci, I. 2026, A&A, 714, A87
  57. [60]Pastore, M. & Calcagni, A. 2019, Frontiers in Physiology, 10, 1089
  58. [61]Raaijmakers, G., Greif, S. K., Hebeler, K., et al. 2021, ApJL, 918, L29
  59. [63]Raithel, C. A. & Most, E. R. 2023b, Phys. Rev. D, 108, 023010
  60. [64]Ravenhall, D. G., Pethick, C. J., & Lattimer, J. M. 1983, Nucl. Phys. A, 407, 571
  61. [65]Ray, P. S. et al. 2019, STROBE-X: X-ray Timing and Spectroscopy on Dynamical Timescales from Microseconds to Years, NASA Probe-class mission concept study report
  62. [66]Riley, T. E. et al. 2019, ApJL, 887, L21
  63. [67]Riley, T. E. et al. 2021, ApJL, 918, L27
  64. [68]Roca-Maza, X., Paar, N., & Colò, G. 2015, J. Phys. G, 42, 034033
  65. [69]Rousseeuw, P. J. & Croux, C. 1993, Journal of the American Statistical Association, 88, 1273
  66. [70]Russotto, P. et al. 2011, Phys. Lett. B, 697, 471
  67. [71]Russotto, P. et al. 2016, Phys. Rev. C, 94, 034608
  68. [72]Samuelsson, L. & Andersson, N. 2007, MNRAS, 374, 256
  69. [73]Scurto, L., Pais, H., & Gulminelli, F. 2024, Phys. Rev. D, 109, 103015
  70. [74]Strohmayer, T. E. & Watts, A. L. 2006, ApJ, 653, 593
  71. [75]Tews, I. 2017, Phys. Rev. C, 95, 015803
  72. [76]Tsang, C. Y., Tsang, M. B., Lynch, W. G., Kumar, R., & Horowitz, C. J. 2024, Nat. Astron., 8, 328
  73. [77]Tsang, D., Read, J. S., Hinderer, T., Piro, A. L., & Bondarescu, R. 2012, Phys. Rev. Lett., 108, 011102
  74. [78]Tsang, M. B. et al. 2009, Phys. Rev. Lett., 102, 122701
  75. [79]Wang, M., Huang, W. J., Kondev, F. G., Audi, G., & Naimi, S. 2021, Chin. Phys. C, 45, 030003
  76. [80]Weih, L. R., Most, E. R., & Rezzolla, L. 2019, ApJ, 881, 73
  77. [81]Zhang, Z. & Chen, L.-W. 2015, Phys. Rev. C, 92, 031301

Paper details

vlnc:W_94hmpwks
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents