This library is a Rust code generator, generating the mathematics you need for a geometric algebra library.
I made it mostly to teach myself the fundamentals of PGA.
Consider it an experiment.
The Geometric Algebra course at Siggraph 2019 https://bivector.net/index.html
Foundation of Game Engine Development, Volume 1, by Eric Lengyel http://terathon.com/blog/projective-geometric-algebra-done-right/ http://terathon.com/pga_lengyel.pdf
Vector names: X Y Z W
Grammar: X²=1 Y²=1 Z²=1 W²=0
Types:
| Name | Blades | Alternative | Interpretation |
|---|---|---|---|
| Vec3 | X Y Z |
direction | |
| Vec4 | X Y Z W |
homogenous point | |
| Point3 | X Y Z W=1 |
normalized point | |
| Line3 | WX WY WZ YZ ZX XY |
direction + moment of a Plücker line | |
| Plane | YZW ZXW XYW ZYX |
!(X Y Z W) | normal + offset |
| Translator3 | YZ ZX XY XYZW |
translation primitive | |
| Rotor3 | WX WY WZ XYZW |
rotation primitive, a.k.a. quaternion | |
| Motor3 | WX WY WZ XYZW YZW ZXW XYW S |
translator + rotor, a.k.a. dual quaternion |
From the above definition, this library generates all the operations that can be done on these types. For instance, it will autmatically realize that Point3 ^ Point3 -> Line3 (wedging two points gives the line that goes through those points) and Plane V Line3 -> Vec4 (the antiwedge of a plane and a line is the point where the plane and line interesect).
The generated code uses newtypes for all vectors and blades, so that x = y; wouldn't compile (since x and y coordinates run along different vectors).
You first specify the the algebra by the signs of the product of the base vectors, e.g. 0++ for 2d projective geometric algebra, which would be the base vectors e0²=0, e1²=1, e2²=1.
You can also give your base vectors more descriptive names, e.g. X/Y/Z/W for your standard homogeneous 3D PGA.
From these all products are generated, creating bivectors (XY, YZ etc), trivectors (XYZ etc), and so on. Together with the Real type they make up the blades of the system. All values are a linear combination of the blades, e.g. 0.5 + 2*X - 42*XZ.
We can give names to groups of blades, for instance naming Vec3={X,Y,Z}, Point3={X,Y,Z,W}, Line3={WX, WY, WZ, YZ, ZX, XY} (Plücker coordinates), Plane={YZW,ZXW,XYW,ZYX} etc.
Note that a value can have multiple types. For instance, in the example above, any Vec3 or also a Point3 with W=0 (an infinite point). The types thus form a Venn diagram.
These types will be combined against each other for all operations, unary (like the dual) as well as binary (multiplication, dot product, wedge, regressive, ...). The generator will notice what dimensions (blades) will be the output, and deduce a type name form that. For instance, Point3 ^ Point3 -> Line3 (wedging two points gives you the line that goes through both points) or Plane V Line3 -> Point3 (the antiwedge of a plane and a line is the point where the plane and line interesect).
As a programmer, my view of Geometric Algebra is as a type safe superset of linear algebra that unifies many differents parts of the standard 3D programming toolset into one theory. Using GA we can combine vectors, points, plücker lines, planes, translators, rotors (quaternions) and motors (dual quaternions) into one framework. This library generates the code for these primitves and all valid operations you can do using them.
Geometric Algebra (GA) is great and very general. We will focus first on the interesting subset of 3D Projective Geometric Algebra (PGA). This is where you denote point and vectors with an extra projective dimension w and define w=0 as directions and w=1 as cartesian coordinates. Thus a vector is expressed as [x, y, z, w].
In textbook geometric algebra, the base vectors are given the names e1/e2/e3/e4 (or sometimes e0/e1/e2/e3). Due to this unfamiliarity and inconsistency I prefer to simply rename them to the familiar names X/Y/Z/W.
So, I use this notations:
2D PGA:
| Class | Blades | Description | |
|---|---|---|---|
| Scalar | S | Numbers | The set of real numbers |
| Vectors | X Y W | Directions | Orthogonal dimensions in projective space |
| Bivectors | YW WX XY | Lines | Orthogonal planes with normals X Y W |
| Trivector | XYW | Area |
I will use lower case letters for x,y,z,w for values (e.g. 3.14) and upper case letters X,Y,Z,W for the basis vectors, which can be thought of as the type of the value.
Consider two arrows on the ground, both with heir names engraved on them: e and n, and they happen to be pointing east and north. What can you do with two arrows?
e ^
------> |
n|
|
One thing you can do with them, is to stick them together at their bases. We call this wedging:
^
|
n|
+------>
e
In geometric algebra, we write this as n^e. If one arrow has length 2 and the other length 3 we get 2N^3E = 6N^E = 6NE`. This represents the plane spanned by the two vectors, with the magnitude representing the area of the parallelogram spanned by the vectors.
The value x means "How much along the X axis?".
XY: How much area do I have in the plane spanned by XY plane, i.e. the plane pointing along the positive Z axis.
Say you have three orthogonal vectors which are the sides of a cube. What is the volume of them together?
Classical thinking would get you |v0|*|v1|*|v2|. In GA, you instead think of the dimensions involved.
Each vector has three scalars: {x: X, y: Y, z: Z} (w=0 since these are directions).
We want to get to the volume, which has dimension XYZ. We want to from low dimension (directions) to a higher one (volume), so what do we do? We wedge them together! Wedging is joining many lower-dimensions things (lines) into one higher dimension things (volume). So the answer is v0^v1^v2. And then you are done!