How Neutron Star Radii Encode the Dense-Matter Equation of State and Hadron-Quark Transition

The paper investigates how future high-precision neutron star (NS) radius measurements encode microscopic information about the dense-matter equation of state (EOS), focusing on a possible first-order hadron–quark phase transition and the resulting mass–radius topology. Within a Bayesian framework using meta-model EOSs with nine microscopic parameters, the authors analyze mock radius measurements R 1.4 = 11.9 ± σR km with σR = 0.9 and 0.1 km for canonical NSs. They introduce inverse EOS–radius mappings that give the posterior mean of each EOS parameter as a function of R 1.4. Their slope measures radius sensitivity, while their curvature determines the leading precision dependence of the posterior mean through the Jensen expansion. Resolving the mappings into four mass–radius topologies, Connected, Disconnected, Both, and No-Quark-Matter, reveals a clear hierarchy of information. The symmetry-energy parameters L (slope) and K sym (curvature) are strongly encoded in R 1.4 and their posterior means shift appreciably with improved radius precision, whereas the higher-order hadronic parameters show stronger topology dependence. Among the transition parameters, the transition density ρ t is the most strongly encoded in R 1.4, while the energy-density jump and quark-matter sound speed are more strongly associated with the topology of the full mass–radius sequence. Since the different topologies have strongly overlapping R 1.4 distributions, even precise radius measurements cannot by themselves identify the topology or uniquely determine the high-density transition properties. These results provide a parameter-dependent hierarchy for assessing the scientific return of future high-precision radius measurements and complementary probes of high-density NS EOS.

The paper begins by noting that recent advances in X-ray observations by NICER and gravitational-wave detections by LIGO/Virgo have improved knowledge about the EOS of neutron-rich matter. With next-generation X-ray timing missions and third-generation gravitational-wave detectors, the uncertainty in NS radius measurements is expected to improve by almost an order of magnitude over the coming decade. The central question addressed is: How is microscopic dense-matter physics encoded into measurable NS radii?

The authors use a meta-model EOS that combines a flexible parameterization of the hadronic EOS with a first-order hadron–quark phase transition to quark matter with a constant speed of sound (CSS). The hadronic EOS is characterized by empirical parameters of symmetric nuclear matter and the symmetry energy, including K 0, J 0, K sym, J sym, L, and E sym (ρ 0), while the transition is characterized by the transition density ρ t, the energy-density jump ∆ε, and the quark-matter sound speed c 2s. The prior ranges for these parameters are given in Table 1. The TOV equations are solved for each EOS parameter set to obtain the mass-radius sequence, and the likelihood includes filters for crust-core transition pressure positivity, thermodynamic stability, causality, and the ability to support NSs at least as massive as 1.97 M⊙.

The inverse EOS–radius mapping is defined as 〈θ i 〉(R) = ∫ θ i P (θ i R) d θ i, which gives the posterior mean of an EOS parameter for a given inferred stellar radius after marginalizing over all remaining parameters. This contrasts with forward modeling via TOV equations. The slope of the mapping indicates how sensitive the inferred parameter is to variations in the radius data, while the curvature determines how the inferred parameter changes as observational precision improves. The Jensen expansion shows that for a smooth mapping θ i = f i (R), the leading finite-precision correction is determined by the curvature f i ′′ (R̄): a convex mapping gives a positive shift in the posterior mean, whereas a concave mapping gives a negative shift.

For the hadronic EOS parameters, the inverse mappings of L and K sym show close similarity between the two radius uncertainties, with both exhibiting substantial correlations with R 1.4. K sym displays a nearly monotonic increase with increasing R 1.4, while L shows a nonlinear but generally increasing behavior. This is consistent with the connection between the pressure of neutron-rich matter around (1 − 2)ρ 0 and the radius of a canonical NS. The topology-resolved mappings for L and K sym largely overlap among the Connected, Both, Disconnected, and No-Quark-Matter categories, indicating that R 1.4 constrains the effective stiffness of the hadronic EOS around 2ρ 0 without uniquely determining the topology.

In contrast, the higher-order hadronic parameters J 0 and J sym show qualitatively different behavior. Their inverse mappings are substantially less monotonic and exhibit visibly different behaviors among the topologies, especially in the lower and intermediate radius ranges. This reflects that J 0 and J sym influence the continuation of the hadronic EOS to higher densities, including the stiffness of matter immediately before the onset of deconfinement, which is crucial for determining whether the appearance of quark matter destabilizes the stellar sequence. The topology therefore acts as an additional discriminator of high-density EOS parameter space not available from the canonical radius alone.

For the hadron–quark transition parameters, the transition density ρ t displays a pronounced and non-monotonic dependence on R 1.4, starting from small radii, increasing substantially to a broad maximum around R 1.4 ≃ 11.5–12.0 km, then decreasing gradually toward larger radii. This non-monotonic behavior demonstrates that the canonical radius does not provide a globally one-to-one mapping onto the transition density. Nevertheless, within the radius interval most strongly favored by the mock data, ρ t changes appreciably with R 1.4, making it substantially more accessible to radius measurements. Reducing σR from 0.9 to 0.1 km reduces the posterior uncertainty and localizes the inferred transition density.

The normalized energy-density jump ∆ε/ε t behaves differently. Its inverse mapping rises from relatively small values at the smallest radii and then becomes approximately flat over a broad interval from roughly R 1.4 ≃ 12 km to 13.5 km. The topology-resolved mapping shows that different mass–radius topologies occupy distinctly different ranges of ∆ε/ε t: the Disconnected category is concentrated near very large energy-density jumps, whereas the Connected category extends toward substantially smaller values, with the Both category in between. This separation persists even though their canonical radii overlap strongly, demonstrating that ∆ε/ε t is primarily associated with the global topology of the stellar sequence rather than with the canonical radius itself.

The quark-matter sound speed c 2s shows an even more striking behavior. Its inverse mapping decreases rapidly from large values at the smallest radii and then becomes nearly flat at c 2s / c 2 ≃ 0.5–0.6 over a wide range of R 1.4. The topology-resolved mapping reveals a strong separation among the classes: the Connected category preferentially occupies a region of relatively large c 2s, while the Disconnected category is concentrated at substantially smaller values, with the Both category in between. This reflects that a relatively stiff quark phase can compensate for the softening associated with the phase transition and allow a Connected sequence to persist, while a strong softening combined with a small quark-matter sound speed can destabilize the star and favor a disconnected hybrid branch.

The posterior PDFs of the transition parameters confirm these findings. For ρ t, reducing σR from 0.9 to 0.1 km shifts the means systematically toward larger values for the Both and Connected categories: the mean ρ t /ρ 0 changes from 4.48 ± 0.87 to 4.84 ± 0.72 for Both and from 4.06 ± 0.77 to 4.31 ± 0.81 for Connected. For ∆ε/ε t, the means are nearly unchanged within each topology: 0.559 ± 0.218 to 0.558 ± 0.225 for Both, 0.761 ± 0.170 to 0.760 ± 0.167 for No-Quark-Matter, and 0.955 ± 0.034 to 0.966 ± 0.028 for Disconnected, with a somewhat larger but still moderate change for Connected (0.511 ± 0.223 to 0.462 ± 0.210). For c 2s / c 2, the means change only modestly with σR within a given topology: for Connected EOSs the mean changes only from 0.561 ± 0.277 to 0.597 ± 0.279, while for Disconnected EOSs it decreases from 0.134 ± 0.203 to 0.093 ± 0.093.

The paper concludes that the gain from improving radius precision is intrinsically parameter dependent. High-precision measurements of R 1.4 most directly improve the inference of the radius-setting parameters L and K sym and provide useful additional information on the transition density ρ t. By contrast, the strength of the transition (∆ε/ε t) and the stiffness of the quark phase (c 2s / c 2) remain more strongly tied to the global mass–radius topology. Since the topology-resolved posterior distributions of R 1.4 overlap strongly, a precise canonical radius cannot by itself identify whether the sequence is Connected, Disconnected, Both, or No-Quark-Matter, nor can it uniquely determine the high-density transition properties. Additional observations sensitive to the global high-density stellar structure are required to distinguish the topologies and constrain the strength and post-transition stiffness of a possible hadron–quark transition.

Improvements to AI Systems:

1. Uncertainty-Aware Inverse Mapping for Parameter Inference

• Implement a Bayesian neural network or Gaussian process that learns the inverse EOS–radius mapping (posterior mean of EOS parameters as a function of R 1.4) directly from simulated mass–radius data.

• Use the slope and curvature of this learned mapping to automatically quantify how sensitive each EOS parameter is to radius precision, and to predict how posterior means shift with improved measurement uncertainty (via the Jensen expansion).

• The improved AI can provide real-time, parameter-specific forecasts of scientific return for any proposed radius measurement precision, without re-running expensive TOV solves.

2. Topology-Aware Classification and Conditional Inference

• Train a classifier (e.g., a deep neural network or random forest) on the four mass–radius topologies (Connected, Disconnected, Both, No-Quark-Matter) using EOS parameters and mock radius data.

• Combine this classifier with a conditional generative model (e.g., a normalizing flow) that predicts EOS parameters given both R 1.4 and a predicted topology.

• The improved AI can:

• Identify which topology is most consistent with a given radius measurement, while explicitly quantifying the overlap in R 1.4 distributions across topologies.

• Provide topology-conditioned posterior distributions for transition parameters (e.g., rho t, epsilon/epsilon t, c s 2/c squared) that are otherwise poorly constrained by radius alone.

3. Precision-Dependent Hierarchical Forecasting

• Develop a meta-learning framework that, given a target radius uncertainty (e.g., sigma R = 0.1 km vs. 0.9 km), predicts which EOS parameters will gain the most information (e.g., reduction in posterior variance or shift in mean).

• Use the paper’s findings to encode a prior hierarchy: L and K sym are most sensitive to R 1.4, followed by rho t, while epsilon/epsilon t and c s 2/c squared are topology-dominated.

• The improved AI can automatically recommend optimal observational strategies (e.g., which radius measurements to prioritize) and flag cases where radius data alone are insufficient, suggesting complementary probes (e.g., tidal deformability, maximum mass) for topology disambiguation.

4. Non-Monotonic Mapping Handling

• Since the inverse mapping for rho t is non-monotonic in R 1.4, implement a multi-modal posterior inference method (e.g., mixture density networks or Bayesian mixture models) that can represent multiple plausible rho t values for a given radius.

• The improved AI can output a full posterior distribution (not just a mean) for transition density, correctly capturing the ambiguity and avoiding misleading single-point estimates.

5. Transferable Emulator for Multi-Messenger Data Fusion

• Build a surrogate model (e.g., a neural network emulator) that maps EOS parameters to both radius and other observables (e.g., tidal deformability, maximum mass) using the same meta-model EOS.

• Train the AI to jointly infer EOS parameters and topology from combined mock datasets (radius + gravitational-wave + X-ray), leveraging the paper’s insight that radius alone cannot break topology degeneracies.

• The improved AI can perform simultaneous multi-observable inference, automatically weighting each probe based on its information content for each EOS parameter and topology class.

6. Active Learning for Next-Generation Observations

• Implement an active learning loop where the AI proposes the most informative next measurement (e.g., radius of a specific NS mass, or a gravitational-wave event) to maximally reduce uncertainty in the most impactful EOS parameters (e.g., L, K sym, rho t) while also attempting to disambiguate topologies.

• The improved AI can optimize the scientific yield of upcoming missions (e.g., NICER follow-ups, LIGO/Virgo upgrades) by prioritizing observations that target the parameter-dependent hierarchy identified in the paper.

We investigate how future high-precision neutron star (NS) radius measurements encode microscopic information about the dense-matter equation of state (EOS), focusing on a possible first-order hadron--quark phase transition and the resulting mass--radius topology. Within a Bayesian framework using meta-model EOSs with nine microscopic parameters, we analyze mock radius measurements R 1.4=11.9 plus or minus sigma R km with sigma R=0.9 and 0.1 km for canonical NSs. We introduce inverse EOS--radius mappings that give the posterior mean of each EOS parameter as a function of R 1.4. Their slope measures radius sensitivity, while their curvature determines the leading precision dependence of the posterior mean through the Jensen expansion. Resolving the mappings into four mass--radius topologies, Connected, Disconnected, Both, and No-Quark-Matter, reveals a clear hierarchy of information. The symmetry-energy parameters L (slope) and K sym (curvature) are strongly encoded in R 1.4 and their posterior means shift appreciably with improved radius precision, whereas the higher-order hadronic parameters show stronger topology dependence. Among the transition parameters, the transition density rho t is the most strongly encoded in R 1.4, while the energy-density jump and quark-matter sound speed are more strongly associated with the topology of the full mass--radius sequence. Since the different topologies have strongly overlapping R 1.4 distributions, even precise radius measurements cannot by themselves identify the topology or uniquely determine the high-density transition properties. These results provide a parameter-dependent hierarchy for assessing the scientific return of future high-precision radius measurements and complementary probes of high-density