nvalchemi-data-structures

Graph-based atomic data models enable efficient representation and batching of atomic systems for GPU workflows. They construct AtomicData and Batch objects, enforcing a Pydantic schema for positions and atomic_numbers while supporting optional energy and cell fields.

Batching multiple atomic graphs for high-throughput simulations is done by creating Batch objects from a list of AtomicData instances. This prepares the combined graph data for device-aware operations and GPU-accelerated pipelines.

Yes, you can construct atomic graph data from ASE Atoms objects. The data structures support ASE integration, allowing you to easily ingest external structures into AtomicData instances for your GPU workflows.

The data schema enforced for atomic data models is Pydantic-based. It requires core fields like positions and atomic_numbers, supports optional system fields such as energy, cell, and pbc, and allows extending data via an info dict.

You need graph-based representations for atomic systems when performing scalable materials simulations, nanostructures analysis, or building machine-learning potentials that require efficient data batching and GPU-accelerated computation.