A robust and generalized library for Geometric Algebra in Rust.
Geometric algebra is an extension of linear algebra where vectors are extended with a multiplication operation called the geometric product, which produces new kinds of operations and vector-like objects that are useful in simplifying and generalizing a number of tricky systems in math and physics.
For instance:
- In physics, using "bivectors" geometric algebra provides a consistent and clean way of representing angular velocities in any dimension, which normally isn't possible in the standard system of angles and vectors.
- Geometric algebra unifies the rotational actions of Complex numbers and Quaternions into a single object called a Rotor that works in all dimensions.
- Using "unit blades", geometric algebra gives a compact and efficient way to represent vector and affine spaces that's easier to work with than sets of basis vectors.
For more information and an in depth explanation, I highly recommend Geometric Algebra for Computer Science by Leo Durst, Daniel Frontijne, and Stephann Mann. A lot of inspiration for this library comes from that book, and the concepts are explained in a way that's easy to follow and visual without falling too deep into overly abstract math.
This crate is split up into a number of modules each representing a different interpretation of the underlying algebra:
algebra
contains the most raw form of Geometric algebra defined purely algebraically. This is primarily used as the core all the other systems are built upon, but it is useful as well for various physical quantities (like position and angular velocity) and abstract mathematical objectssubspace
interprets the structs inalgebra
as weighted vector spaces and orthogonal transformation (like rotations and reflections) in ℝnhomogenous
(planned) interprets the structs inalgebra
as affine spaces and their operations in ℝn-1 using homogeneous coordinates
Additionally, the module wedged::base
contains the core components common
to all subsystems.
Naming in geometric algebra can vary slightly across authors, so the particular conventions have been chosen:
- in
wedged
, aSimpleBlade
refers to an object constructed from the wedge product of vectors whereas aBlade
refers to any object of a particular grade - a
Rotor
is used exclusively for objects with unit norm that encode rotations whereasEven
is used for a general object from the even subalgebra Multivector
s refer to a general element in the algebra- The operators
*
/Mul
,^
/BitXor
, and%
/Rem
are used for the geometric, wedge, and generalized dot products respectively
Additionally, most structs in this crate support indexing where each component
is ordered according to a particular basis convention chosen for this crate.
See the docs on BasisBlade
For more information.
In order to generalize code for different scalar types and dimensions,
wedged
makes heavy use of generics. The system is based upon the
system in nalgebra
, and, to increase interoperability, reuses a lot of its
components.
The system is designed such that a number of different levels of abstraction are possible:
use wedged::algebra::*;
let v1 = Vec3::new(1.0, 0.0, 0.0);
let v2 = Vec3::new(0.0, 1.0, 0.0);
let plane = v1 ^ v2;
let cross = plane.dual();
assert_eq!(plane, BiVec3::new(0.0,0.0,1.0));
assert_eq!(cross, Vec3::new(0.0,0.0,1.0));
When not using generics, things act much like you would expect from a vector library.
use wedged::algebra::*;
use wedged::base::ops::*;
fn midpoint<T:Clone+ClosedAdd+ClosedDiv+One>(v1:Vec3<T>, v2:Vec3<T>) -> Vec3<T> {
let two = T::one() + T::one();
(v1 + v2) / two
}
fn cross<T:RefRing>(v1:Vec3<T>, v2:Vec3<T>) -> Vec3<T> {
(v1 ^ v2).dual()
}
let v1 = Vec3::new(1.0, 0.0, 0.0);
let v2 = Vec3::new(0.0, 1.0, 0.0);
assert_eq!(cross(v1,v2), Vec3::new(0.0,0.0,1.0));
assert_eq!(midpoint(v1,v2), Vec3::new(0.5,0.5,0.0));
let v1 = Vec3::new(1.0f32, 0.0, 0.0);
let v2 = Vec3::new(0.0f32, 1.0, 0.0);
assert_eq!(cross(v1,v2), Vec3::new(0.0f32,0.0,1.0));
assert_eq!(midpoint(v1,v2), Vec3::new(0.5f32,0.5,0.0));
For generic scalars, wedged
structs implement traits and functions based on
what their scalars implement. For instance:
- Structural traits like
Index
,AsRef<[T]>
, andBorrow<[T]>
are always implemented - Basic traits like
Clone
,fmt
traits,PartialEq
,Eq
,Hash
, andDefault
are all implemented if the scalar implements them Copy
is implemented if the scalar isCopy
the struct isn't aDynamic
dimension- Additive operations like
Add
,Sub
, andNeg
as well as scalar operations likeMul
andDiv
. are implemented if the scalar has them. The assign variants and operations between references are implemented as well if the scalars implement them. - Traits for the geometric (
Mul
), wedge (BitXor
), and dot (Rem
) products are implemented if the scalar has addition, subtraction, negation, zero and multiplication between references - anything involving a norm, constructing a rotation, etc, requires
nalgebra::real_field
Div
usually has the same requirements asMul
, but is only implemented on a few specific structs (likeRotor
andVersor
)
In order to make managing these requirements easier, a number of trait aliases
are included in wedged::base::ops
that combine common groupings of operations
together
use wedged::base::*;
use wedged::algebra::*;
fn point_velocity<T,N:Dim>(p:VecN<T,N>, angular_vel: BiVecN<T,N>) -> VecN<T,N> where
T:AllocBlade<N,U1>+AllocBlade<N,U2>+RefRing
{
p % angular_vel
}
let p = Vec2::new(2.0,0.0);
let av = BiVec2::new(3.0);
assert_eq!(point_velocity(p, av), Vec2::new(0.0,6.0));
let p = Vec3::new(0.0,2.0,0.0);
let av = BiVec3::new(3.0,0.0,0.0);
assert_eq!(point_velocity(p, av), Vec3::new(0.0,0.0,6.0));
For a generic dimension, for every wedged
struct
used in the function or trait, the scalar must have the corresponding Alloc*
trait bound for that struct. These are all included in wedged::base::alloc
,
and there is one for each subset of the geometric algebra.
For example, for a scalar T
and dimension N
, using a VecN
requires T:AllocBlade<N,U1>
bound, using an Even
needs a T:AllocEven<N>
bound, etc.
Finally, even if a non-generic scalar is used, these are still required inside a where clause on the function or trait.
use wedged::base::*;
use wedged::algebra::*;
fn project<T,N:Dim,G:Dim>(v:VecN<T,N>, b:Blade<T,N,G>) -> VecN<T,N> where
U1: DimSymSub<G>,
DimSymDiff<U1,G>: DimSymSub<G,Output=U1>,
T:AllocBlade<N,U1> + AllocBlade<N,G> + AllocBlade<N,DimSymDiff<U1,G>> + RefDivRing
{
let l = b.norm_sqrd();
(v % &b) % b.reverse()/l
}
let v = Vec3::new(1.0, 2.0, 0.0);
let line = Vec3::new(0.0, 1.0, 0.0);
let plane = BiVec3::new(0.0, 1.0, 0.0);
assert_eq!(project(v,line), Vec3::new(0.0, 2.0, 0.0));
assert_eq!(project(v,plane), Vec3::new(1.0, 0.0, 0.0));
Using a generic grade follows nearly the same rules as for generic dimensions but with an added generic for the grade. However, with grade altering operations like the wedge or dot products are used, additional bounds are required to allow the compiler to add or subtract the grades of the two blades.
For instance:
- The wedge
^
product requiresG1: DimAdd<G2>
- The dot
%
product requiresG1: DimSymSub<G2>