利用点估计和后验分布量化中子星物态方程（EOS）的不确定性
Quantifying uncertainty in the neutron-star equation of state using point estimates and posterior distributions
浏览论文内容
中文总结 AI 辅助
该研究对比点估计与分布推断方法，结合多类信息开展中子星EOS不确定性量化，发现确定性MLP在高密度区不确定性带更窄，提出需用表征条件概率分布的方法实现可靠高密度EOS不确定性量化。
中文摘要 AI 辅助
我们采用相同的切比雪夫（Chebyshev）和分段线性参数化方法，对比点估计与分布推断两种方法，开展中子星物态方程（EOS）的不确定性量化研究。我们将中子星质量-半径关系、引力波潮汐形变信息，融合到贝叶斯、多层感知机（MLP）及归一化流（normalizing flow）框架中。尽管各方法给出的EOS平均行为相近，但确定性MLP在高密度区域生成的不确定性带显著更窄。我们表明，该特性与点估计目标相关——其会将简并解映射到条件均值，而非完整参数后验分布；相比之下，归一化流生成的分布与贝叶斯推断结果更一致。研究结果表明，要实现高密度EOS的可靠不确定性量化，需采用能表征条件概率分布而非仅点估计的方法。
英文摘要
We investigate uncertainty quantification for the neutron-star equation of state (EOS) by comparing point-estimation and distributional inference approaches using the same Chebyshev and piecewise-linear parameterizations. We combine neutron-star mass--radius and gravitational-wave tidal-deformability information within Bayesian, multilayer-perceptron (MLP), and normalizing-flow frameworks. Although the methods yield similar mean EOS behavior, the deterministic MLP produces substantially narrower uncertainty bands at high densities. We show that this behavior is associated with the point-estimation objective, which maps degenerate solutions toward the conditional mean rather than representing the full parameter posterior. By contrast, the normalizing flow yields distributions more consistent with the Bayesian inference. Our results demonstrate that reliable uncertainty quantification of the high-density EOS requires methods that represent conditional probability distributions rather than only point estimates.
