Tensors #4

First off, hello there. As it currently stands, I’ve been up for more than 20 hours, in an attempt to straighten out my sleep schedule once and for all (this is probably futile, but I can only hope not). I’m bored, have no energy to move, and certainly am in no state to learn, but require something to keep me from sleeping, so as a consequence, you get the present post, in which I will mindlessly regurgitate the procedure for constructing the tensor product of two vector spaces (this is therefore the much-awaited “existence proof” which I procrastinated due to its finicky nature and pronouncèd lack of aesthetic appeal). I apologize in advance if I manage to botch anything, since I probably won’t read over it until I’ve had a good sleep or something. Anyways, here goes nothing.

§5. Construction of the tensor product (as a quotient of a free vector space)

Suppose as usual that we fix a field \mathbb{F} and we are given two \mathbb{F}-vector spaces \mathsf{V} and \mathsf{W}. In order to “construct a tensor product”, we must do two things: first of all, construct the space \mathsf{T} := \mathsf{V} \otimes \mathsf{W} itself, exhibit a suitable bilinear mapping \otimes : \mathsf{V} \times \mathsf{W} \to \mathsf{T}, and prove that the universal property is satisfied.

The construction will go as follows: we will take the free vector space on \mathsf{V} \times \mathsf{W}, denoted C(\mathsf{V} \times \mathsf{W}), define a certain “special” subspace N, and then quotient it out. The mapping \otimes will then be given simply by \pi, the canonical projection (“reduction mod N“).

So let’s do this! Let N be the subspace of C(\mathsf{V} \times \mathsf{W}) generated by the vectors (\lambda x_1 + \mu x_2, y) - \lambda(x_1,y) - \mu(x_2, y) and (x, \lambda y_1 + \mu y_2) - \lambda (x,y_1) - \mu (x,y_2), where x, x_1, x_2 \in \mathsf{V}, y, y_1, y_2 \in \mathsf{W}, and \lambda, \mu \in \mathbb{F}. Now then, let us define \mathsf{T} to be the quotient C(\mathsf{V} \times \mathsf{W}) / N. Defining the map \otimes : \mathsf{V} \times \mathsf{W} \to \mathsf{T} by putting v \otimes w := \pi(v,w) where \pi : C(\mathsf{V} \times \mathsf{W}) \to \mathsf{T} is the canonical projection (or “quotient map”), we can observe bilinearity quite easily. Since every vector of the form (x, \lambda y_1 + \mu y_2) differs from \lambda(x, y_1) + \mu(x, y_2) only by an element of N, we observe that

x \otimes (\lambda y_1 + \mu y_2) = \pi(x, \lambda y_1 + \mu y_2) = \lambda \pi(x, y_1) + \mu \pi(x, y_2) = \lambda( x \otimes y_1 ) + \mu( x \otimes y_2)

thereby demonstrating that \otimes is linear in its second argument. In an identical fashion it is shown that \otimes is linear in its first argument. Therefore \otimes is a bilinear map. It now remains to check that (T1) and (T2) are satisfied.

To verify (T1), we need simply show that the vectors v \otimes w generate \mathsf{T}. Let z \in \mathsf{T} be arbitrary. We know this element must be the equivalence class of some element of C(\mathsf{V} \times \mathsf{W}) (note the direct analogue between subspaces and equivalence relations when we talk about quotient spaces). Hence (due to the general form of such an element) we see that

\displaystyle z = \pi \sum_{\alpha, \beta} \lambda_{\alpha \beta} (v_\alpha, w_\beta) = \sum_{\alpha, \beta} \lambda_{\alpha \beta} \pi(v_\alpha, w_\beta) = \sum_{\alpha, \beta} \lambda_{\alpha \beta} (v_\alpha \otimes w_\beta).

To verify (T2), suppose we are given a bilinear map f : \mathsf{V} \times \mathsf{W} \to \mathsf{Z}, where \mathsf{Z} is some vector space. Note that the set of all pairs (x,y) for x \in \mathsf{V} and y \in \mathsf{W} form a basis for the space C(\mathsf{V} \times \mathsf{W}), therefore this is enough to uniquely determine a linear map \psi : C(\mathsf{V} \times \mathsf{W}) \to \mathsf{Z} such that \psi(x,y) = f(x,y). In order to see that this map induces our desired map \varphi : \mathsf{T} \to \mathsf{Z}, we must show that the subspace N is sitting inside the kernel of \psi. For example, suppose we have some vector of the form

z = (\lambda x_1 + \mu x_2, y) - \lambda(x_1, y) - \mu(x_2, y).

Then we observe that by the definition of \psi and the bilinearity of f,

\psi(z) = \psi(\lambda x_1 + \mu x_2, y) - \lambda \psi(x_1, y) - \mu \psi(x_2, y) = f(0, y) = 0.

A similar phenomenon occurs if z = (x, \lambda y_1 + \mu y_2) - \lambda(x,y_1) - \mu(x,y_2). Therefore indeed N \subseteq \ker \psi, and we therefore know by the universal property of quotient spaces that \psi induces a linear map \varphi from the quotient space (tensor product) \mathsf{T}, into \mathsf{Z}, such that \varphi \circ \pi = \psi and hence \varphi \circ \otimes = f thereby proving (T2). This completes the construction.


About mlbaker

just another guy trying to make the diagrams commute.
This entry was posted in abstract algebra, linear algebra and tagged , , , . Bookmark the permalink.

One Response to Tensors #4

  1. William says:

    20 hrs? Oh dear.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s