AMDAT: An Open-Source Molecular Dynamics Analysis Toolkit for Supercooled Liquids, Glass-Forming Materials, and Complex Fluids

AMDAT (Amorphous Molecular Dynamics Analysis Toolkit) is an open-source C++ toolkit for post-processing molecular dynamics trajectories, focused on high-performance static and dynamic analyses of amorphous, glassy, and polymer materials, including supercooled liquids and complex fluids. In this paper, we describe AMDAT’s design for efficient long-timescale analysis via in-memory trajectory handling and exponential time sampling, and we demonstrate representative workflows for widely used observables such as radial distribution functions, structure factors, intermediate scattering functions, and neighbor correlations.

💡 Research Summary

The manuscript introduces AMDAT (Amorphous Molecular Dynamics Analysis Toolkit), an open‑source C++ library designed specifically for post‑processing molecular dynamics (MD) trajectories of non‑crystalline systems such as supercooled liquids, glass‑forming materials, polymers, and other complex fluids. The authors argue that while MD simulation engines have become highly mature and widely shared, the ecosystem of analysis tools has lagged behind, especially for soft‑matter and glass‑forming applications where subtle structural and dynamical correlations are crucial. Existing analysis packages are often fragmented, lack thorough validation, and force researchers to re‑implement basic routines, hampering reproducibility.

AMDAT addresses this gap through a set of deliberate design choices. First, the entire trajectory is read into memory at initialization. This “in‑memory” approach eliminates repeated file I/O, enabling near‑instantaneous access to any frame and dramatically accelerating cross‑frame calculations such as two‑time correlation functions (e.g., mean‑square displacement, intermediate scattering functions). The memory overhead is modest—typically two to three times the raw trajectory size—but is justified by the speed gains for long‑time analyses.

Second, AMDAT employs exponential time sampling. Users can request frames that are densely spaced at short times and progressively coarser at longer times, allowing a single analysis run to span many orders of magnitude in physical time without generating prohibitively large trajectory files. The authors demonstrate that this strategy reduces both storage and computational cost while preserving the fidelity of time‑dependent observables.

Third, the toolkit is built around a modular, object‑oriented data model. Core abstractions include “trajectory list” (a collection of particle trajectories), “trajectory bin list” (spatially binned subsets), “multibody list” (groups of atoms forming molecules, clusters, or strings), “value list” (scalar per‑particle, per‑frame quantities such as displacement magnitude), and “neighbor list” (distance‑based or Voronoi‑based neighbor information). These objects can be combined, filtered, and passed between analysis routines, enabling flexible construction of complex workflows. For example, a value list of particle mobilities can be thresholded to generate a new trajectory list of “mobile” particles, which can then be examined for clustering or spatial heterogeneity.

Fourth, AMDAT provides a scriptable, non‑interactive command‑line interface. Input scripts resemble LAMMPS syntax and support loops, conditionals, and variable evaluation, allowing users to encode entire analysis pipelines in a reproducible, human‑readable form. The toolkit reads standard trajectory formats (LAMMPS dump, XYZ, GROMACS XTC) and also permits custom column mapping so that auxiliary per‑atom data (velocities, forces, energies) can be incorporated directly into value lists. Output is written as plain‑text, tab‑delimited files or modified trajectory files compatible with downstream visualization tools such as Python, MATLAB, OVITO, or VMD.

The functional repertoire of AMDAT includes well‑tested implementations of radial distribution functions (RDF), static structure factors S(q), intermediate scattering functions F_s(k,t), mean‑square displacement (MSD), neighbor decorrelation, persistent neighbor tracking, radius of gyration, and various orientational correlation functions. These routines have been refined over more than fifteen years within the authors’ research group and have been employed in dozens of published studies, providing confidence in their reliability and scientific accuracy.

The paper details the internal architecture: a top‑level “System” object handles trajectory loading, metadata (box size, time scheme), and provides looping mechanisms over frames. “Trajectory” objects store per‑particle coordinate time series, while specialized subclasses (AtomTrajectory, MoleculeTrajectory) capture atomic or molecular entities. The modular data objects are composable; analysis functions accept one or more objects as input and return new objects, facilitating pipeline construction where the output of one step becomes the input of the next.

A representative workflow is presented: the user defines the system and composition, creates a trajectory list for a specific species, computes per‑particle displacements (value list), thresholds to select the most mobile 5 % of particles, builds a multibody list of these mobile clusters, and finally calculates a cluster size distribution and spatial correlation functions. All steps are expressed in a concise script, demonstrating the toolkit’s ability to automate complex, reproducible analyses.

Performance benchmarks show that, for a 10⁶‑particle, 10⁶‑step supercooled liquid simulation, AMDAT reduces total analysis time by more than 80 % compared with conventional file‑based scripts, and the exponential time sampling enables accurate intermediate scattering functions covering 10⁴‑fold time ranges with a single run.

In conclusion, AMDAT fills a critical niche in the MD community by delivering a high‑performance, open‑source analysis platform tailored to amorphous and glass‑forming materials. Its in‑memory design, exponential time sampling, modular data structures, and scriptable workflow collectively provide speed, flexibility, and reproducibility that were previously lacking. The authors intend to extend the toolkit with additional observables (e.g., nonlinear response functions), GPU acceleration, and a Python API, fostering broader adoption and community‑driven development.