Bridging molecular dynamics and electronic-structure calculations with SIEVE: a graph-based approach

Bridging molecular dynamics and electronic-structure calculations with SIEVE: a graph-based approach C. Picarelli 1, G. Raffaini 1, and M. Tommasini 1 1 Dipartimento di Chimica, Materiali e Ingegneria Chimica “G. Natta”, Politecnico di Milano, Piazza Leonardo da Vinci 32 – 20133 Milano (Italy) email: [email protected] Multiscale modeling is instrumental for connecting atomistic processes to experimentally observable properties. SIEVE is an open-source C++ framework designed to automate the extraction and analysis of statistically meaningful molecular microstates from molecular dynamics (MD) trajectories, providing a scalable and reproducible bridge between dynamical simulations and electronic-structure calculations. SIEVE infers molecular connectivity from Cartesian coordinates and atomic numbers using graph-theoretical methods, without requiring predefined bonding information. Its core capabilities include graph-based isomorphism detection [1], adaptive bond recognition, and efficient I/O handling through the high-performance H5MD format [2], enabling rapid identification, classification, and indexing of species in large-scale simulations with tens of thousands of atoms. Recent enhancements introduce radial distribution functions and datadriven adaptive solvation shell definitions, ensuring physically consistent and representative sampling of molecular configurations across solvated systems and aggregates. By systematically selecting representative microstates, SIEVE automates the generation of inputs for high-level electronic-structure calculations, particularly Density Functional Theory (DFT), thereby supporting a data-driven multiscale workflow for predicting spectroscopic observables such as UV–Vis absorption spectra. Three representative applications illustrate its effectiveness: characterization of molecular aggregation events and clustering statistics along MD trajectories; extraction of solvated microstates for TD-DFT simulations of UV–Vis absorption, rationalizing the role of solvation and conformational effects; and analysis of intermolecular connectivity through colored molecular graphs, providing detailed statistics on local environments and neighborhood distributions. SIEVE thus offers a versatile and automated tool for multiscale computational spectroscopy, combining efficiency, reproducibility, and rigorous physical accuracy with minimal user intervention. [1] – McKay, Brendan D., and Adolfo Piperno. "Practical graph isomorphism, II." Journal of symbolic computation 60 (2014): 94-112. [2] – de Buyl, Pierre, Peter H. Colberg, and Felix Höfling. "H5MD: A structured, efficient, and portable file format for molecular data." Computer Physics Communications 185.6 (2014): 1546-1553.