Dynamic axis management: add/remove/reshape axes at specification time #2
Description
Support declarative addition and removal of axes in the system specification, so users can straightforwardly extend a model with new trait, spatial, or structured dimensions.
Motivation
A TRIDENT-like model starts with trait axes (phytoplankton size θ, zooplankton size ϑ) and a spatial axis (depth z). A user should be able to:
- Add a new trait axis (e.g. light-harvesting pigment ratio) with one YAML block
- The compiler should automatically:
- Extend the state tensor dimensions
- Wire diffusion operators for the new axis
- Adjust reaction terms that depend on the new trait
- Validate that rate functions have the right arity
Tasks
-
Axistype hierarchy:TraitAxis,SpatialAxis,StructuredAxis(age, stage, etc.) - Each axis declares: name, bounds, default resolution, diffusion coefficient (optional), boundary conditions
-
add_axis!(spec, axis)— extend a compiled SystemSpec with a new dimension -
remove_axis!(spec, name)— project out a dimension (integrate or marginalise) - Validate: reaction terms must accept the correct number of trait arguments after axis changes
- Tensor-product grid assembly: the full state is the outer product of all axis grids
- Resolution independence: axes can have different resolutions; the compiler handles indexing
Cross-references
- OpEngine.jl Dynamic axis management: add/remove/reshape axes at specification time #2: the adaptive grid refinement consumes axis declarations from here
- OpEngine.jl Core YAML-to-callable system compiler #1: the solver must handle variable-dimension state vectors
- trident Core YAML-to-callable system compiler #1: TRIDENT is the first consumer — 3 trait axes + 1 spatial axis
- ACCIDDA/op_system: Python equivalent should get a parallel issue
Boundary
This is specification — how axes are declared and validated. The grid points (uniform or adaptive) are computed by OpEngine.jl. The evaluation of whether adding an axis improves predictions is ModelCriticism.jl.