Reproducible engines and interfaces on heterogeneous compute resources for structural analysis of deformed systems

We present updates to d-SEAMS: Deferred Structural Elucidation Analysis For Molecular Simulations; a high performance library with a backend in C++ and a Lua frontend. d-SEAMS was initially developed with an eye for ice nucleation studies and continues to have a strong focus on ring structure determination and the analysis of coverage metrics for identifying phase transitions. Recent studies have expanded its usage into the realm of clathrate studies, which involved a restructuring of the internal data structures. To better integrate with heterogeneous computing resources offered with Python bindings, the newer code structure has been augmented to have Python bindings. These enable further workflows which were previously complicated for users to design and run. Following the strategy and design of larger codebases like PyTorch, d-SEAMS v2 provides granular access to the primitives and has helper functions to adapt external codes like ASE for generalization. We discuss the newer design, along with the algorithmic updates through a series of documented systems which are being considered for publication elsewhere. A new plugin framework is also designed to allow easier access from external codes as well. This release takes the engine and system closer to the original goal of being an umbrella software with specialized "glue-code" bindings for both reproducible results (by storing the YAML configurations) and also interactive usage (through Python or Lua bindings).