Tensor Fields and Simplicial Complexes
Listening to Freya Holmér last night I started to get glimmers of an idea I had long ago about how to represent vector spaces in computational processes using this recursive abstract type:
of {getx : 'a vector,
diff : 'a point -> 'a point,
move : 'a point -> 'a point,
scale : 'a -> 'a point,
proj : 'a point -> 'a}
with
let fun self x = POINT
move = fn (POINT pr) =>
diff = fn (POINT pr) =>
scale = fn i =>
proj = fn (POINT pr) =>
end
fun getx (POINT pr) = #getx pr
fun diff (POINT pr) = #diff pr
fun move (POINT pr) = #move pr
fun scale (POINT pr) = #scale pr
fun proj (POINT pr) = #proj pr
end
Then we can use points to instantiate vectors of 'a points starting with the unit type for a 0-simplex.
I think this datatype will also represent his box arithmetic:
Here's his playlist on box arithmetic.
My comment:
38:11 I like this description a lot, because it seems to connect quite well with the idea of vectors and covectors. If I were trying to implement msets on a computer I would use vectors of multiplicities based on some enumeration of the previous level of mset. So for msets of natural numbers it would be the obvious enumeration 0, 1, 2,... The problem is for msets of msets of naturals there is no obvious finite enumeration. For example, before you present the multiplicity of [2] you would want to have presented that of each mset of msets of naturals before [2], but there are infinitely many so this is not a well-ordering. But maybe there is a smarter way to do it that I haven't thought of. Dana Scott came up with a neat enumeration of recursive enumeration operators, so these things are not always as impossible as they might at first look. [See Scott's 2014 paper Stochastic λ-calculi: An extended abstract which still makes my head spin!] This might be related to the basis theorem and the axiom of choice. See the basis theorem in nlab for a good introduction.
He just posted this:
Subscribe to Norman Wildberger.
So the idea is to start with simplicial complexes rather than fixed vectors. In a sense, that is what vector spaces already are: we never have actual data in any physical space which define what the basis vectors of the space really are, they are just orthonormal vectors in some unspecified units, and in a Tensor field we don't even know what orthogonal really means either. So the angle and length scales are a matter of convention, not of "physically evidential" fact. Then on the basis of just a simplicial complex we can define a metric tensor field. That would be much more like space as we experience it, I think.
See Gabriele Carcassi on The Correspondence Between Quantum and Classical Mechanics
This is the previous talk he mentions:
On the physical significance of the existence of a well-ordering on the reals, see basis theorem in nlab:
The original proof of the existence of Hamel bases (after which the concept was named) was for the case of the real numbers regarded as a rational vector space and used (not directly Zorn's lemma but) the well-ordering theorem.
They are running an Assumptions of Physics Summer School, according to this post. See Assumptions of Physics - Summer School 2026 - June 24-26.
See Christoph Benzmüller - Many Logics, One Methodology for some ideas about how automated theorem provers can be used in such an exercise.
Subscribe to Gabriele Carcassi.
Norman Wildberger thinks Algebraic Calculus is the way forward.
Subscribe to Norman Wildberger.
See About Logic with Seunghyun Song and Jordi Fairhurst for more on vector spaces.
Toby made a great video about preorders today:
Subscribe to Tibees.
I guess that means I am supposed to learn Lean now:
My comment:
In Eilenberg's 1976 book "Automata, Languages and Machines" he defines Relations, Monoids and Categories and then writes (on the last page I have access to) "An example of a category is the category R of relations. The objects are sets and the morphisms in R(X, Y) are all the relations f: X → Y, with composition being composition of relations." There is a copy of volume II on the Internet Archive but you aren't allowed to look inside,... It's a three-volume work intended to lay a rigorous mathematical foundation for a chunk of computer science. I'm sure you know about it.
A finite-dimensional vector space is isomorphic to its dual, but only if you pick a basis. It's also isomorphic to its double dual, and that isomorphism requires no choices at all. Mathematicians called the second kind "natural" long before anyone could say what that meant precisely. In 1945, Samuel Eilenberg and Saunders Mac Lane invented an entire branch of mathematics to make the distinction rigorous. The framework they built, consisting of categories, functors, and natural transformations, turned out to capture something far more general than its origin. This video builds category theory from the ground up: categories as the minimal axioms for composition, functors as structure-preserving translations between mathematical domains, natural transformations as the maps that commute with everything, universal properties as the principle that what a construction does matters more than what it's made of, and adjunctions as the paired constructions that tie it all together.
Subscribe to Computable Secrets.
Subscribe to The Hidden Library.
Subscribe to More in Depth.
Frederic Schuller's lecture on tensor spaces:
1:27:54 Vectors and matrices are just witch mathematics!
1:55:37 On the importance of the order in which things are presented: first without the basis, then with the basis to get the dimension, then back to basis-free representation, then introducing the components, ... If you start out thinking of a vector as a list of numbers then all sorts of confusions arise.
2:00:52 On n-forms and determinants. I am sure there is a way to understand this in terms of simplicial complexes.
Tangent vector spaces from smooth manifolds:
TL;DR: Position vectors are just pointless confusion!
See Frederic Schuller on Metric and Topological Spaces and Frederic Schuller on Topological Manifolds and Manifold Bundles.
41:30 The exterior derivative, without choosing any basis and so without components.
See the full series of 28 lectures: The Geometric Anatomy of Theoretical Physics.
Subscribe to Frederic Schuller.
Norman Wildberger's idea is to synthesize smooth functions from power-series representations
Subscribe to Wild Egg Maths.
Now I just need to understand how these meshes of control points can be put on co-ordinate-free simplicial complexes and instantiated under some top-form.
Marketing! See Gabriele Carcassi - A Simple Explanation of Quantum Mechanics and Frederic Schuller on Metric and Topological Spaces.
Comments
Post a Comment