Skip to main content
22 events
when toggle format what by license comment
Sep 29, 2021 at 0:50 comment added Zach Teitler I’m a little taken aback. There are several ways that people think about tensors, I listed a few.
Sep 28, 2021 at 23:55 comment added David Roberts @LSpice because I recognise the physicists notion of tensor [field] when I see it, because that was my first education. Also, when does someone say 'coordinates' instead of 'basis' in the purely algebraic setting that Will is trying to tease out?
Sep 28, 2021 at 23:53 comment added Will Sawin @LSpice In my original comment I intended to use a theorem like "To produce a functor from a groupoid to another category, it suffices to fix the value of that functor on one object in each equivalence class together with its automorphisms." Since you mentioned based vector spaces i thought it might be helpful to mention the category of based vector spaces with vector space isomorphisms and explain why it is relevant here. (Though I guess, among other things, I should probably have said why the category of based vector spaces with basis-preserving maps is not relevant.)
Sep 28, 2021 at 23:44 comment added LSpice @WillSawin, I guess this should be the last I say, because it's clear I'm misunderstanding. You didn't say based vector spaces at all; I did (and I meant the more restricted class of morphisms, though I didn't say so). Did you mean to? Or is your point that it's irrelevant?
Sep 28, 2021 at 23:42 comment added Will Sawin @LSpice: I didn't say the category of finite-dimensional based vector spaces and vector space isomorphisms respecting the choice of basis! I said the category of finite-dimensional based vector spaces and vector space isomorphisms. So there is no need to respect the choice of basis.
Sep 28, 2021 at 23:40 comment added LSpice @WillSawin, I'm sorry to be so obtuse, but I think that that cannot be part of an equivalence of categories. The functor needs to assign different isomorphisms to the two isomorphisms $1, -1 : k \to k$ to be part of an equivalence of categories; but two different isomorphisms cannot both respect the choice of a basis of the 1D vector space $k$.
Sep 28, 2021 at 23:29 comment added Will Sawin @LSpice Just write down all the vector spaces in a list and choose a basis for each, applying the axiom of choice. (Without the axiom of choice you need to use anafunctors to make this make sense.)
Sep 28, 2021 at 23:13 comment added LSpice @DavidRoberts, how do you know that @‍WillSawin's suggestion isn't what was meant?
Sep 28, 2021 at 23:11 comment added LSpice isomorphisms. You refer to "the $\operatorname{GL}_n$ representation"; is it possible that you intended one of the categories to carry the additional datum of a $\operatorname{GL}_n$ representation in some way?
Sep 28, 2021 at 23:11 comment added LSpice @WillSawin, I am sorry; I don't understand how these categories are equivalent. To exhibit an equivalence involves in particular a functor from the category of vector spaces with isomorphisms, to the category of based vector spaces with isomorphisms. I do not think it is possible functorially to pick a basis of a vector space, even if we require only that the functor respect
Sep 28, 2021 at 22:44 comment added David Roberts @Will that may be, but it's not what was meant.
Sep 28, 2021 at 22:13 comment added Will Sawin @LSpice A perhaps slightly different approach would be to define a tensor as a function from the set of bases of the vector space to the set of multidimensional arrays of numbers that satisfies some transformation rules.
Sep 28, 2021 at 22:11 comment added Will Sawin @LSpice If two categories are equivalent, a functor from one (up to isomorphism) is the same data as a functor from the other. The category of finite-dimensional vector spaces and vector space isomorphisms is equivalent to the category of based finite-dimensional vector spaces and vector space isomorphisms. In concrete terms, you just pick a set of coordinates arbitrarily and define tensors using those coordinates, and the $GL_n$ representation tells you how to deal with changes of coordinates.
Sep 28, 2021 at 21:18 comment added LSpice @WillSawin, shouldn't the objects of your category be based vector spaces? I may be misunderstanding, but precisely the fact that there are multiple choices of basis for a vector space seems to prevent me from thinking of (5) as a definition in the way you propose: if tensors are specified in coördinates, how to attach one to a non-coördinatised vector space?
Sep 28, 2021 at 14:16 comment added Will Sawin @DavidRoberts If you interpret "change of coordinates" as "automorphisms of $V_1,\dots, V_n$", (5) seems to me like a perfectly reasonable definition of a tensor. It's equivalent, I would say, to defining the tensor product as a functor from the $n$-fold product of the category of finite-dimensional vector spaces and isomorphisms, by observing that each object in this category is isomorphic to $F^m$ for some $m$ and defining an explicit representation of $GL_{m_1} (F) \times \dots \times GL_{m_n}(F)$.
Sep 28, 2021 at 5:54 comment added Zach Teitler @DavidRoberts I've edited my answer and I hope that it is less bothersome now.
Sep 28, 2021 at 5:52 history edited Zach Teitler CC BY-SA 4.0
added 112 characters in body
Sep 28, 2021 at 5:44 comment added David Roberts The problem is, a tensor in the sense of 1-4 is an entirely different object to 5. OK, so I've got a span of some basic tensors. how does this transform under "changes of coordinates". Really the issue is that physicists are slack (and I come from that culture...) and say 'tensor' when they mean 'tensor field', or 'tensor-valued function', and give unhelpful definitions like "A tensor is a thing that transforms like a tensor". Mathematically, we have separate words for these things...
Sep 28, 2021 at 4:13 comment added Zach Teitler I concede that I'm blurring lines. The lists for vector spaces, and for bundles, would have been mostly the same, so I kind of mashed them together. It's true, I'm abusing the ambiguity of the term "tensor" by importing a notion that people use for bundles into the vector spaces setting. Nevertheless it's an idea that some people use, and surely that makes a profound difference in how they think of tensors, compared to (1)-(4).
Sep 28, 2021 at 0:18 comment added David Roberts 5 is different to the others, since it is talking about sections of tensor powers of a vector bundle and their coordinates with respect to local frames. The others are purely algebraic. So something like 'tensor-valued functions'. No one confuses a real number (Cauchy seq or Dedekind cut or...) with a function.
S Sep 28, 2021 at 0:09 history answered Zach Teitler CC BY-SA 4.0
S Sep 28, 2021 at 0:09 history made wiki Post Made Community Wiki by Zach Teitler