AI 中文总结
针对稠环等构建非冗余内坐标困难的问题，提出SONIC方法，在SMITH中实现，能确定性构建广义内坐标，经多种验证，输出含多种信息的冻结坐标合约，为相关研究提供了有效工具。
AI 中文摘要
内坐标是分子结构的自然语言，但对于稠环、高配位、分子对称性和弱束缚片段，自动构建非冗余内坐标仍很困难。我们提出了SONIC（面向对称性的非冗余内坐标），它在SMITH中实现，是一种对局部化、对称性适配的广义内坐标的确定性构建。SMITH接受冻结的分子状态、提供的拓扑或基元，或通过修订固定的独立感知内核输入笛卡尔坐标。它对普通基元进行分类，添加化学类型的特殊和复合族，保护环和片段坐标，并使用解析威尔逊B行进行块局部秩约简。局部伪对称性对中心、环和键域进行排序，同时在均匀块内应用精确的分子点群投影。输出是一个冻结坐标合约，具有人类可读的分解、对称性和秩诊断、解析一阶导数以及可选的高斯16序列化。优化、内坐标到笛卡尔坐标的实现、力场和高阶海森矩阵变换仍为后续任务。验证涵盖无环、稠合、桥连、高配位非共价和eta3金属配合物情况，包括旋转/置换不变性、Python/ Fortran检查和独立的广义内坐标评估。
英文摘要
Internal coordinates are the natural language of molecular structure, but automatic non-redundant constructions remain difficult for fused rings, high coordination, molecular symmetry, and weakly bound fragments. We present SONIC (Symmetry-Oriented Non-redundant Internal Coordinates), implemented in SMITH, as a deterministic construction of localized, symmetry-adapted generalized internal coordinates. SMITH accepts a frozen molecular state, supplied topology or primitives, or Cartesian input through a revision-pinned standalone perception kernel. It classifies ordinary primitives, adds chemically typed special and composite families, protects ring and fragment coordinates, and performs block-local rank reduction using analytic Wilson B rows. Local pseudosymmetry orders center, ring, and bond domains, while exact molecular point-group projection is applied within homogeneous blocks. The output is a frozen coordinate contract with human-readable decompositions, symmetry and rank diagnostics, analytic first derivatives, and optional Gaussian 16 serialization. Optimization, internal-to-Cartesian realization, force fields, and higher-order Hessian transformations remain downstream responsibilities. Validation covers acyclic, fused, bridged, high-coordinate, non-covalent, and eta3 metal-complex cases, including rotation/permutation invariance, Python/Fortran checks, and independent GIC evaluation.
Comments64 pages, 6 figures; Supporting Information supplied as a 7-page ancillary PDF; code and reproducibility materials are described in the manuscript
