发表机构
University of Amsterdam; Advanced Research Center for Nanolithography（阿姆斯特丹大学; 纳米光刻先进研究中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出Vitriflow方法，将熔淬分子动力学中的隐式选择转化为显式决策链，通过数值稳定性、描述符校准、伪影筛选和统计收敛，生成可重复的非晶系综，并在a-SiO₂、a-Si₃N₄和a-Sm₂O₃中验证了其有效性。
AI 中文摘要
熔淬分子动力学广泛用于构建非晶材料模型，但所得系综由通常隐含的选择定义：数值设置、熔融温度、液态保持时间、淬火速率、系统尺寸和生成后筛选。我们引入vitriflow，一种将这些选择转化为显式决策链的计算材料学方法。该框架在材料特定的描述符空间中耦合了数值稳定性、基于描述符的协议校准、用户定义的伪影筛选以及生成分析系综的统计收敛性。我们在a-SiO$_2$、a-Si$_3$N$_4$和a-Sm$_2$O$_3$上演示了该方法，分别测试了四面体网络保真度、异极性氮化物的MG2 $\rightarrow$ PBE $\rightarrow$ HSE06 DFT精修以及混合配位稀土氧化物中的非晶/晶体区分。vitriflow能够从氧桥缺陷二氧化硅中分离出无缺陷二氧化硅，量化常见a-Si$_3$N$_4$结构群体中的DFT精修响应，并在不施加固定配位的情况下移除再结晶的Sm$_2$O$_3$结构。结果是一种可重复的生成非晶系综的路径，其数值设置、热协议、筛选操作和统计精度根据材料问题选择，而非先验假设。
英文摘要
Many structure-property relations in amorphous materials are encoded not only in the mean structure, but in the prevalence, correlations and spatial organisation of minority environments across length scales. Yet atomistic models are commonly validated against bulk averages and a nominal number of structures, without establishing whether the finite population resolves the motifs invoked to explain material behaviour. We present an auditable framework, implemented in the Vitriflow software package, that connects calibrated melt-quench generation to explicit populations and quantity-specific uncertainty. This approach is demonstrated on three distinct benchmark systems. In a-SiO$_2$, strain is followed from intra-tetrahedral deformation through compressed Si--O--Si bridges to primitive 3-ring topology, showing that the lower angular tail is an organised medium-range population rather than undifferentiated variation about the network mean. In matched a-Si$_3$N$_4$ refinements, DFT redistributes density and first-neighbour length scale, sharpens the force-field first shell and repairs marginal contacts, while the two DFT descendants retain essentially the same first-neighbour topology. In a-Sm$_2$O$_3$, the amount and connectivity of partially ordered domains are quantified as they emerge from the disordered parent population, while the mixed coordination characteristic of the amorphous oxide remains robust. Mean structure, distribution tails, inherited topology and connected order are therefore distinct materials variables with distinct uncertainties. There is no universal ensemble size: the Vitriflow package records what was generated, which population was analysed and what that finite population establishes, providing a reproducible route from amorphous structure datasets to quantitatively supported structure-property hypotheses.
