NumpyEinsumOperator is the default operator type Unroled sparse operators would only make sense for large batch sizes, and high dimensional products

Generate algebra by associating bit-patterns of integers with blade patterns This implementation uses vectorized numpy code, and implements the com

Register methods that dynamically dispatch according to the subspace of a multivector

Multi-Linear symbolic operator

Plain sparse container type it is not yet bound to a product type knowledge of how to actually multiply is left to Operator classes

Implement product using JAX einsum. This seems to provide a nice balance between compilation speed and runtime speed, though sparse evaluation seems

Implement multilinear operator using einsum on dense kernel in numpy This is the best implementation in numpy for small subspaces or unbatched multi

Implement product using einsum.

Coordinate Matrix to be used as a construction helper

All the boilerplate bookkeeping which is generic to any particular implementation of the algebra's logic

matrix with a specific operator and bound operand subspace given knowledge of what product and subspace we are multiplying with, we can 'write our G

Caches the creation of concrete operator instances from abstract operator instances NOTE: this is more a warehouse than a factory. also, whats the

A simple point-to-point constraint between rigid bodies I can say from experience that more complex joints can be implemented in GA with comparative

matrix container bound to an unbound operator

Groups together some convenience functionality to allocate concrete MultiVector types

Contains constructors for a rich set of geometric linear operators, and acts as a cache of those operators. Note that the operators merely manage t

Helper objects to extend .at[].set syntax of jax arrays to multivectors and operators

Describes a subspace of a full geometric algebra, by means of an array of bit-blades, denoting the blades present in the subspace

Namespace for subspace constructors, and a place to provide caching / flyweight mechanism

Interface to wrap an abstract linear operator, with a concrete implementation in a specific framework Abstract in the sense that this should not be

Abstract context object; to be overridden by backend specific logic

Abstract multivector type; To be used as a base class for implementations in specific backends Note that there is essentially no GA-logic in this c

rates are encoded as a bivector

single grade that squares to a scalar

Anchors are a pair of points, describing the location of the constraint in the body local frame

Register methods that dynamically dispatch according to static type-like attributes of its arguments This is an abstract base class; specific imple

like a motor that does not obey m*~m==1

Implement product using plain broadcasting in JAX. This is identical to the approach used for quaternion multiplication in alphafold. The JAX compil

Concrete realization of abstract type in JAX

Unroll a (generally sparse) kernel into computations on individual nonzero entries This will give very good performance in jax in high dimension, d

Dual to 2-vec; momenta, forques note; questionable to call momenta a line in higher dims, if it cant be factored into the join of two points in gen

Positions/rotation states are encoded as even-graded multivectors

Concrete realization of abstract type in numpy

Unroll a (generally sparse) kernel into computations on individual nonzero entries For small batches or small subspaces, this is unlikely to be the

dual to 1-vec. encodes positions

quick and dirty python codegen example

Concrete realization of abstract type in JAX

Unroll a (generally sparse) kernel into computations on individual nonzero entries

All functionality that can be implemented with just access to the blades and algebra Note: both the subspace and operator class implement this inter

Implement product using plain broadcasting.

Concrete realization of abstract type in torch

Unroll a (generally sparse) kernel into computations on individual nonzero entries