AI 中文总结
该研究通过KDE-HDR方法将349个脉冲星glitch样本分为紧凑、扩展两类，发现平均反向等待时间是核心预测因子，揭示了两类glitch的物理机制差异。
AI 中文摘要
脉冲星 glitch 表现出多样的行为，通常被归类为“船帆座型”或“蟹状星云型”，但这些定性类别无法提供可重复的定量边界。我们基于反向等待时间Δt₋与 glitch 振幅分数Δν/ν的联合分布构建了客观分类方案，通过交叉匹配焦德雷尔班克天文台和 ATNF glitch 数据库，汇编了来自22颗脉冲星的349个 glitch 样本。对每个源，我们使用二维核密度估计（KDE）结合最高密度区域（HDR）推导几何描述符，包括 HDR 面积、各向异性比和 HDR 外占比。对 KDE-HDR 描述符的层次聚类揭示了两类与视觉检查无关的形态类别：紧凑类（Compact class）对应小 HDR 面积、高各向异性和有限的外围占比；扩展类（Extended class）对应更宽的 HDR 占比和更强的低密度延伸。我们通过自助重采样和超参数敏感性测试确认了该划分对大多数源的有效性，但发现两颗高产脉冲星（PSR J0537-6910 和 PSR J1341-6220）处于中间位置。逻辑回归将平均反向等待时间⟨Δt₋⟩识别为最强的类别预测因子（样本内 AUC=0.88；留一法交叉验证 AUC_CV=0.59，反映样本量有限）。此外，使用Δt₊的平行分析恢复了相同的两类结构，但稳定性降低。数据表明，紧凑类对应近完全的储层耗尽，而扩展类反映部分雪崩式释放，因此对扩展类脉冲星G的短期测量可能低估真实的地壳超流体分数。
英文摘要
Pulsar glitches display diverse behaviors. They are commonly labeled as "Vela-like" or "Crab-like," yet these qualitative categories do not provide reproducible quantitative boundaries. We constructed an objective classification scheme based on the joint distribution of backward waiting time $Δt_{-}$ and fractional glitch amplitude $Δν/ν$. A sample of 349 glitches from 22 pulsars was compiled by cross-matching the Jodrell Bank and ATNF glitch databases. For each source, we used two-dimensional kernel density estimation (KDE) together with highest-density regions (HDRs) to derive geometric descriptors, including the HDR area, anisotropy ratio, and out-of-HDR fraction. Hierarchical clustering of the KDE--HDR descriptors revealed two morphology classes independent of visual inspection. In this convention, the Compact class is associated with small HDR areas, elevated anisotropy, and limited peripheral occupancy. The Extended class is associated with broader HDR occupancy and stronger low-density extensions. We found that bootstrap resampling and hyperparameter sensitivity tests confirmed this partition for most sources. Nevertheless, two prolific pulsars (PSR J0537-6910 and PSR J1341-6220) were found to occupy intermediate positions. Logistic regression identified the mean backward waiting time $\langleΔt_{-}\rangle$ as the strongest class predictor (in-sample $\mathrm{AUC}=0.88$; leave-one-out $\mathrm{AUC}_{\mathrm{CV}}=0.59$, reflecting the limited sample size). Additionally, a parallel analysis using $Δt_{+}$ recovered the same two-class structure but with reduced stability. The data suggest that the Compact class is consistent with near-complete reservoir depletion, whereas the Extended class reflects partial, avalanche-like releases. As a result, short-term measurements of $G$ in Extended-class pulsars may underestimate the true crustal superfluid fraction.
Comments22 pages, 6 figures, 7 tables; accepted for publication in The Astrophysical Journal
