Sleipnir.jl

Sleipnir.jl is the foundational package of the ODINN ecosystem, providing the core data structures and infrastructure on which all other packages are built. Every other ODINN package (Huginn, Muninn, ODINN) depends on Sleipnir, and each re-exports its symbols so downstream users rarely need to import Sleipnir directly.

Sleipnir defines the glacier geometry and climate data containers (Glacier2D, Climate2D), the simulation parameter hierarchy (Parameters, SimulationParameters, PhysicalParameters), the law abstraction used to plug physical or machine-learning computations into the PDE solvers (Law, AbstractLaw), and the results container (Results). It also hosts the cache infrastructure (MatrixCache, ScalarCache, and their NoVJP variants), which is central both to memory-efficient forward simulation and to the VJP computations used by inverse modelling workflows.

Data for Sleipnir is preprocessed by the Python package Gungnir. Glacier objects are assembled with initialize_glaciers(), which reads the stored NetCDF files via Rasters.jl. For glaciers already covered by the ODINN Hugging Face dataset the data is downloaded automatically at precompilation, so running Gungnir yourself is only needed for new glaciers or custom climate sources — see Glaciers for both paths and for the Overrides.toml mechanism that points ODINN at a custom directory.

Use directly vs. use ODINN.jl

Use Sleipnir directly when you want to:

• Build or inspect glacier data structures (Glacier2D, Climate2D) without running a full simulation.
• Prototype a new Law type or AbstractInput that will later be used in Huginn or ODINN.
• Write a lightweight script that reads preprocessed glacier data and extracts fields (thickness, surface elevation, climate) without loading the full simulation stack.

Use Huginn when you need to actually run a forward ice flow simulation on those data structures, and ODINN.jl when you need the end-to-end differentiable pipeline (UDE training, classical and functional inversion) — both assemble Sleipnir types into runnable workflows for you.

Minimal usage example

using Sleipnir

# Construct simulation parameters (multiprocessing disabled for local runs)
params = Parameters(
    simulation = SimulationParameters(
        tspan = (2010.0, 2015.0),
        multiprocessing = false,
        use_MB = true
    ),
    physical = PhysicalParameters()
)

# Load pre-initialized glaciers (requires preprocessed data in ~/.ODINN/)
glaciers = initialize_glaciers(["RGI60-11.00897"], params)

glacier = glaciers[1]
@show glacier.rgi_id, glacier.nx, glacier.ny
@show size(glacier.H₀)   # initial ice thickness grid
@show size(glacier.S)    # surface elevation grid

Extending Sleipnir

To add a new Law type, a new dynamic input (AbstractInput), or new observation data to Glacier2D, see the Extending ODINN guide.

API reference

See Sleipnir API for the full list of exported types and functions.