If a computer were able to do Clifford multiplication in a single process, it'd be able to do some NP tasks in polynomial time. Clifford algebras are a sweet intersection of combinatoric and geometric thinking.
Edit: didn't even read the article, just upvoted because of the topic.
Intuitively this makes sense: in the GA formalism a "single" term requires 2^n fields (where n is the spatial dimension), and "simple" operations like addition and multiplication thus require 2^n (or more) operations to evaluate.
If you could instead somehow do those operations in O(1) time you would clearly pick up a rather nice speedup, but again I'm curious what the problems and encodings (as GA) actually look like.
– Starting with standard linear algebra (vector spaces, matrices, etc.) leads to confusion. These ideas are best developed within the framework of geometric algebra, where their features can be easily proved / reasoned about, but only after the basic features of geometric algebra have been presented first. After a bit of exposition, the relationship to more commonly used mathematical formalisms can be explained. Same story for trigonometry, barycentric coordinates, complex numbers, quaternions, projective geometry, differential forms, Pauli and Dirac matrices, Lie theory, representation theory, lattice theory, etc.)
– The geometric product is the basic operation. Inner (“dot”) and outer (“wedge”) products of pairs of vectors are derived products. Deriving other subsidiary products from the basic geometric product for various niche purposes and learning to reason about their various features or discussing how you might use one or another to solve particular problems in various ways is a good exercise.
If a and b are vectors, then by definition a∧b = ½(ab – ba) = –b∧a. (We’ve intentionally cancelled out the commutative parts of the product, leaving us something purely anti-commutative.)
For two vectors, by definition a·b = ½(ab + ba) = b·a. (We’ve intentionally cancelled out the anti-commutative parts of the product, leaving something purely commutative.)
That the geometric product of two vectors can be thus split apart by grade (i.e. ab = a∧b + a·b) is a consequence derived from the basic axioms. It so happens that the purely anti-commuting part of the geometric product of two vectors has the dimension/orientation of the plane they lie in, whereas the purely commuting part has the dimension of a scalar. This just falls right out of the basic definition of the geometric product.
– The initial presentation should not focus on coordinates. Coordinates are an arbitrary choice, not inherent to the geometry.
Ideally lessons should start with a vector space (or affine space) where we assume we know how to add or multiply two vectors (probably the most useful way to start building intuition is to just show vectors as arrows in a picture rather than written down as some explicit expression), and then develop a full set of multivectors on top. Figuring out how to stick some arbitrary coordinate system into place for making concrete computations is then straightforward to understand. But if students get used to working with coordinates, they will generally have great difficulty ditching them to think more abstractly. David Hestenes calls this a “math virus”: http://geocalc.clas.asu.edu/pdf/MathViruses.pdf
If you start with an affine space (e.g. the Euclidean affine plane), you can pick an arbitrary origin point “zero”, and then treat the other points in your space as zero plus some vector displacement, and get to work with the displacement vectors in the derived vector space. You can then pick an arbitrary direction and magnitude to be your e₁ direction, and an arbitrary orientation of the plane (I) or an arbitrary orthogonal e₂ vector of the same magnitude. This lets you keep always in mind that your chosen origin and basis were arbitrary, and can be changed at will to suit the problem.
– This presentation doesn’t discuss much of any of the geometrical relationships between vectors, bivectors, lines, planes, etc., or how you might go about using geometric algebra to solve practical geometry (or Newtonian mechanics, or computer graphics, or whatever) problems.
The difficult part about geometric algebra is not the basic definitions, but rather learning to think with it fluently, and practicing combining its abstractions. I personally don’t feel like I have reached a level of real fluency, but only basic familiarity. Mainly because I habitually reach for more standard tools which I know better from using them for years of schoolwork.
The “sunset geometry” blog post which didn’t get any discussion here a week or two ago was IMO a better introduction. https://news.ycombinator.com/item?id=13162368 http://www.shapeoperator.com/2016/12/12/sunset-geometry/
The article states at the beginning he want's to start with a concrete example, instead of diving into abstractions and formalisms. You respond by complaining this isn't abstracted enough.
I get your point, because this might set up some wrong intuitions. On the other hand though, the more abstract approach might be so dense as to resist any intuition at all. In my experience, wrong intuition is a better starting point for learning math that no intuition at all.
I don’t think we should start with something entirely abstract. Indeed, I want to start with something much more concrete: physical pictures of vectors as arrows, and a collection of concrete geometry problems to solve. If we later want to layer in a coordinate system (e.g. because it helps us solve one of our listed problems), we can draw several possible sets of coordinates over the top of our pictures, and think about how the computations in each coordinate system change while the underlying vector arithmetic stays the same.
Coordinates are an abstraction layered on top of vector spaces, and it is difficult for many students to understand the difference between the system or model being studied and the arbitrary details of one representation. I think focusing on coordinates narrows students viewpoint and leads them to a more rote/computational and less conceptual understanding. To the point that e.g. many undergraduate physics majors have difficulty with tests of basic competence about vector concepts, despite needing to solve a variety of hard computational problems.
Indeed, this specific failing of conventional formalisms is one of the primary reasons to switch to geometric algebra as a modeling tool in the first place.
It’s like trying to teach someone to play chess by looking at “algebraic notation” instead of starting by looking at a board and moving physical pieces around, or teaching someone phonology of a new language by looking at prose descriptions in their original language and static diagrams of a mouth, instead of starting by just directly listening to the sounds.
New languages, systems, abstractions, etc. are best learned through (at least some amount of) immersion in the mode of thinking/communication of the language, not purely through formal analysis using external tools. The presentation in the OP leads me to believe that the author (like myself, to be honest) is not actually fluent with thinking about geometric algebra concepts / tools, but is treating them like a novice second-language learner.
(Sorry if that sounds harsh. I’d love to buy the blog post author a coffee or beer if he’s ever in San Francisco.)
Though I have to admit not to know geometric algebra, from both you and the article, I gather it was created as an improvement on linear algebra. For those less innately interested in maths, starting from such a position tends to work much better than starting from scratch.
I feel like there are 2 methods aimed at different people. For the 'real' mathematicians, the approach you advocate is better suited. For those who use maths only as a tool, and do not so much appreciate the innate properties of maths, the easier to follow (but less informative) method of the author might be better.
Sadly, there is no single 'best method' for teaching a thing.
What do you mean by "in a single process"?
Also see this nice recent blog post http://www.shapeoperator.com/2016/12/12/sunset-geometry/
Lately I've been more interested in the deeper workings of Clifford algebras, which is their intimate connection with quadratic forms. When I get to a computer I'll try and post some resources.