Constrained Gaussian-process bridge prior for neutron-star equation-of-state inference

This paper introduces a novel, model-agnostic method using constrained Gaussian-process bridges to generate stable, causal, and thermodynamically consistent nonparametric priors for neutron-star equation-of-state inference, enabling the flexible integration of diverse theoretical constraints without requiring iterative shooting.

The Gist

The Big Picture: Mapping the Unseen Interior

Imagine a neutron star as a giant, super-dense city. We can see the "skyline" from the outside (we know its mass and size), but we cannot see the buildings inside. The "Equation of State" (EoS) is essentially the blueprint for how the matter inside this city is packed together.

Scientists want to figure out this blueprint. They have some clues from the bottom of the city (low density, like regular atoms) and some clues from the very top (high density, where physics gets weird). But the middle part? That's a mystery.

The problem is that if you try to guess the blueprint randomly, you might draw a building that defies physics (like one that collapses instantly or moves faster than light). Previous methods tried to guess the middle by drawing lines between the clues, but they often got stuck or made bad guesses because they couldn't easily enforce the "laws of physics" while doing so.

The New Method: The "Smart Bridge"

This paper introduces a new way to guess the missing blueprint. The authors call it a "Constrained Gaussian-Process Bridge."

Here is how it works, broken down into three simple steps:

1. Building the Frame (The "Fractal" Scaffold)

Imagine you have two points: a low-density point (A) and a high-density point (B). You need to draw a line connecting them that represents the star's interior.

• Old way: You might try to draw a smooth curve, but it's hard to make sure the curve never breaks the rules of physics.
• This paper's way: They start by drawing a very "noisy," jagged line that zig-zags wildly between A and B. But here's the trick: they only draw the zig-zags inside a specific "safe zone." This safe zone is a 3D volume defined by the laws of physics (causality, stability, and energy conservation).
• The Analogy: Think of this like a fractal tree. You start with a trunk. You add a branch. Then you add smaller branches to that branch, and smaller ones to those. You keep doing this infinitely. The result is a structure that has detail at every single scale, but it is strictly contained within the "safe zone" of the forest. This ensures that every single possible path they generate is physically possible, even if it looks messy.

2. Smoothing the Rough Edges (The "Diffusion" Step)

The jagged, fractal lines from step 1 are too messy to be real stars. They need to be smoothed out, but you can't just blur them like a photo, or you might accidentally blur them outside the safe zone (breaking the laws of physics).

• The Solution: They use a mathematical "heat diffusion" process. Imagine pouring hot water over a rough stone. The heat spreads out, smoothing the surface, but the water stays on the stone.
• The Magic: By carefully controlling how this "heat" spreads, they turn the jagged fractal lines into smooth, realistic curves. Crucially, this smoothing process is designed so that the lines never leave the "safe zone." They stay causal (nothing moves faster than light) and stable.

3. Tuning the "Texture" (The Correlation Length)

One of the coolest features of this method is that the scientists can control how "smooth" or "bumpy" the final blueprint is.

• Short Correlation: The blueprint can change quickly. One layer of the star might be stiff, and the next layer soft. This allows for complex, detailed structures.
• Long Correlation: The blueprint changes slowly. If the star is stiff at the bottom, it tends to stay stiff for a long way up.
• The Analogy: Think of this like clay. You can sculpt the clay to have sharp, jagged edges (short correlation) or smooth, rolling hills (long correlation). The method lets the scientists choose the "texture" of the star's interior without breaking the laws of physics.

What Did They Find?

When they applied this new method to real data from neutron stars (like their masses and sizes measured by telescopes), they found a consistent story:

1. The "Stiffening" Phase: Just above the normal density of atoms, the matter gets very "stiff" (hard to squeeze). This is necessary to support the heavy weight of massive neutron stars.
2. The "Softening" Phase: As you go deeper and denser, the matter starts to "soften" again.
3. The Connection: This pattern—getting stiff then getting soft—happens naturally because of the global rules of physics. It suggests that something interesting might be happening inside the core, perhaps a change in the type of matter (like a phase transition), but the method proves this pattern is a requirement of physics, not just a lucky guess.

Why Is This Better?

• No "Shooting" Needed: Old methods often had to play a game of "guess and check" (shooting) to see if a blueprint worked. This method builds the blueprint so that it always works by construction.
• No Bias: It doesn't assume the star looks like a specific model. It explores all possible shapes that fit the rules.
• Unified: It connects the low-density physics (atoms) and high-density physics (quarks) in one smooth, continuous framework without needing to switch rules halfway through.

In short, the authors built a physics-compliant "3D printer" that can generate infinite possible blueprints for neutron stars, ensuring every single one is physically possible, and then used real data to see which blueprints are the most likely to be true.

Technical Summary

Technical Summary: Constrained Gaussian-Process Bridge Prior for Neutron-Star Equation-of-State Inference

Problem Statement
The inference of the neutron-star (NS) equation of state (EoS) from macroscopic observables relies on constructing priors that are model-agnostic yet physically consistent. Existing nonparametric approaches, such as standard Gaussian processes (GPs), struggle to simultaneously enforce global physical constraints: thermodynamic consistency, mechanical stability, and causality (speed of sound $c_s \leq c$). While parametric models can satisfy these constraints, they often act as biased estimators with uncontrolled correlations. Conversely, methods that incorporate high-density constraints from perturbative quantum chromodynamics (pQCD) and low-density constraints from chiral effective field theory (chiral EFT) often rely on "shooting" procedures, intermediate likelihood functions, or ad hoc switching between regimes. These approaches can introduce dependence on matching points or result in the rejection of most generated EoSs because the constraints are not built directly into the prior generation.

Methodology
The authors propose a new nonparametric prior generation method based on a constrained Gaussian-process (GP) bridge. The method constructs EoSs that connect low-density limits (from chiral EFT) and high-density limits (from pQCD) while strictly adhering to physical constraints by construction. The procedure consists of two main steps:

1. Self-Similar (Fractal) Refinement:
   Instead of starting with white noise, the method generates a space of "maximally noisy" EoSs through a recursive, self-similar refinement process. Starting with boundary conditions defined by thermodynamic triplets $\beta_L = (\mu_L, n_L, p_L)$ and $\beta_H = (\mu_H, n_H, p_H)$, the algorithm iteratively introduces intermediate points $\beta_0$ within the physically allowed volume $V_{\beta_L, \beta_H}$. This volume is defined by the constraints of mechanical stability ($n(\mu)$ is single-valued), causality ($\mu/n \cdot dn/d\mu \geq 1$), and thermodynamic consistency (correct pressure integration). Points are sampled uniformly within this allowed volume, creating EoSs with structure at all density scales but minimal short-range correlations.

2. Diffusive Smoothing (Constrained GP Bridge):
   To introduce local correlations and ensure differentiability, the authors apply a heat-kernel smoothing (diffusion) to the chemical potential $\mu(n)$ as a function of a flow time $\tau$. The evolution is governed by the diffusion equation:
   $$\partial_\tau \mu(n, \tau) = \partial_n [D(n) \partial_n \mu(n, \tau)]$$
   Crucially, the diffusion coefficient $D(n)$ is chosen to preserve the physical constraints:

   • Mechanical Stability: Preserved if $D(n) > 0$.
   • Causality: Preserved if $D'' - D'/n \leq 0$.
   • Thermodynamic Consistency: The process conserves energy density within the domain, with only negligible violations at the boundaries that are mitigated by tapering $D(n)$ or reinjecting energy.

   By selecting $D(n) \propto n^2$, the method achieves a correlation length $\sigma$ that is a fixed fraction of the density ($\sigma/n = c$), allowing for fine structure at low densities while spanning the large range to pQCD densities. This results in a modified GP-bridge construction where the local correlation structure is Gaussian, but the global structure retains the non-trivial anti-correlations required by physical constraints.

Key Contributions

• Unified Framework: The method integrates low-density (chiral EFT) and high-density (pQCD) constraints into a single, unified prior without intermediate likelihoods or shooting procedures.
• Constraint Satisfaction: The construction guarantees that every generated EoS is thermodynamically consistent, mechanically stable, and causal by design, eliminating the need to reject invalid samples.
• Flexible Correlation Control: The diffusion approach allows for the tuning of the sound-speed correlation length, bridging the gap between conservative (short correlation) and theory-informed (long correlation) priors.
• Global Anti-Correlation: The method naturally induces long-range anti-correlations in the speed of sound: EoSs that stiffen at intermediate densities are generically required to soften at higher densities to satisfy causality and pQCD limits.

Results
The authors applied the framework to infer the EoS using updated observational data, including mass measurements of PSR J0348+0432 and PSR J1614−2230, joint mass-radius constraints from NICER (PSR J0740+6620, PSR J0614−3329, PSR J0437−4715), and tidal deformability from GW170817.

• EoS Behavior: The posterior distributions consistently show a stiffening of the EoS immediately beyond the chiral-EFT regime to support $2M_\odot$ stars, followed by a systematic softening at higher densities ($n \gtrsim 1 \text{ fm}^{-3}$).
• Correlation Length Dependence: While the local properties of the EoS (e.g., peak sound speed) depend on the chosen correlation length, the mass-radius relation remains largely robust. However, longer correlation lengths allow the EoS to remain stiff over extended density ranges, slightly increasing the maximum radii.
• Phase Structure Indicators: The posterior samples show a trace anomaly $\Delta$ approaching zero at high densities and a conformal distance $d_c$ decreasing at high densities, consistent with the onset of quark matter. However, a strong first-order phase transition (discontinuity in the derivative of pressure) cannot be excluded.
• Observational Constraints: The global thermodynamic correlations allow individual observations (e.g., PSR J0614−3329) to constrain the EoS at densities far beyond the central densities of the observed stars. The analysis highlights a tension between the radius constraints for PSR J0740+6620 and other measurements.

Significance and Claims
The paper claims that the constrained GP-bridge prior offers a physically motivated, computationally efficient alternative to existing methods. Unlike previous approaches that may exclude physically admissible behaviors or rely on arbitrary matching points, this method fills the entire allowed volume of EoSs without extending beyond it. The authors emphasize that the observed softening at high densities is a natural consequence of the global constraints imposed by causality and pQCD, rather than an artifact of local correlations. They assert that this framework provides a transparent way to tune the sound-speed correlation length and facilitates direct comparison with standard GP priors while retaining the essential non-trivial global structure of the physically allowed EoS space. The authors note that the constraining power of the framework is expected to improve as pQCD calculations reach higher orders (e.g., N3LO) with reduced uncertainties.