Abstract Nonsense

Crushing one theorem at a time

Bilinear Forms


Point of post: In this post I give a quick overview of the very basic concepts in the topic of “bilinear forms”. The main things covered being what a bilinear form on the product of two vector spaces \mathscr{U}\times\mathscr{V} is, and finding a basis for it. Incidental little facts may pop up.

Remark: If \mathscr{U,V} are vector spaces, we follow Roman’s (in the author’s opinion) better notation \mathscr{U}\boxplus\mathscr{V} for the external direct sum and reserve \mathscr{U}\oplus\mathscr{V} for the internal direct sum. Also, we shall use the notation A\leqslant B to denote “A is a linear subspace of B“. Lastly, for k\in\mathbb{N} we define [k]=\left\{1,\cdots,k\right\}

Motivation

We know that given a vector space \mathcal{V} over a field F it is fruitful to study the dual space of \mathscr{V}, denoted as always \text{Hom}\left(\mathscr{V},F\right) or \mathscr{V}^*, which is the set of all linear mappings \varphi:\mathscr{V}\to F.  We have also seen that it is interesting to study \mathscr{U}\boxplus\mathscr{V} given vector spaces \mathscr{U,V} over a field F. It makes sense then to ask questions about maps B:\mathscr{U}\boxplus\mathscr{V}\to F. This kind of mapping has arisen before in our studies, for example the map

f:\mathscr{V}\boxplus\mathscr{V}^{*}\to F:(x,\varphi)\mapsto \varphi(x)

Even more common, is the dot product on usual real coordinate space

\displaystyle \langle\cdot,\cdot\rangle:\mathbb{R}^n\boxplus\mathbb{R}^n\to\mathbb{R}:\left((x_1,\cdots,x_n),(y_1,\cdots,y_n)\right)\mapsto \sum_{j=1}^{n}x_jy_j

Maybe less obvious, but another very important example is the determinant of 2\times 2  complex matrices when thought of as being functions of the column vectors, namely:

\det:\mathbb{C}^2\boxplus\mathbb{C}^2\to\mathbb{C}:\left(\begin{bmatrix}z_1 \\ z_2\end{bmatrix},\begin{bmatrix} z_3\\ z_4\end{bmatrix}\right)\mapsto z_1z_4-z_2z_3

Bilinear Forms

All of these are certainly very important, and commonplace functions the working mathematician encounters on a day-to-day basis. So the question becomes, how do these maps relate, are they specific examples of some larger definable class of maps on vector spaces? Intuition first may tell us that they are linear functionals on the external direct product \mathscr{U}\boxplus\mathscr{V}, but a quick check botches this idea, namely:

\det\left(\begin{bmatrix}1+i\\2+i\end{bmatrix},\begin{bmatrix}3+i \\4+i\end{bmatrix}\right)\ne\det\left(\begin{bmatrix}i \\ i\end{bmatrix},\begin{bmatrix}i\\i\end{bmatrix}\right)+\det\left(\begin{bmatrix} 1 \\ 3\end{bmatrix},\begin{bmatrix} 2 \\4\end{bmatrix}\right)

But, a little bit of digging reveals that while they aren’t linear in both entries simultaneously they are linear in each entry individually. For example, one can easily check that

\det\left(\begin{bmatrix}z_1+z_2\\ z_3+z_4\end{bmatrix},\begin{bmatrix}z_5\\z_6\end{bmatrix}\right)=\det\left(\begin{bmatrix}z_1\\z_3\end{bmatrix},\begin{bmatrix}z_5\\z_6\end{bmatrix}\right)+\det\left(\begin{bmatrix}z_2\\z_4\end{bmatrix},\begin{bmatrix}z_5\\z_6\end{bmatrix}\right)

More generally we call a mapping

B:\mathscr{U}\boxplus\mathscr{V}\to X

(where X is any unspecified vector space over Fbilinear if

B\left(\alpha_1x_1+\alpha_2x_2,y\right)=\alpha_1B(x_1,y)+\alpha_2B(x_2,y)

And

B\left(x,\beta_1y_1+\beta_2y_2\right)=\beta_1B(x,y_1)+\beta_2B(x,y_2)

If it happens that these are mappings into the ground field F, symbolically:

B:\mathscr{U}\boxplus\mathscr{V}\to F

we call B a bilinear form. Somethings become immediately apparent from this definition

Remark: From here on out, unless specified otherwise, we will have that \mathscr{U,V} are vector spaces over a field F and B:\mathscr{U}\boxplus\mathscr{V}\to F is a bilinear form.

Theorem: B(x,0)=B(y,0)=0 for any x\in\mathscr{U} and y\in\mathscr{V}.

Proof: This follows immediately since B(x,0)=B(x,0+0)=B(x,0)+B(x,0), and so B(x,0)=0. Similarly, B(0,y)=B(0+0,y)=B(0,y)+B(0,y) and so B(0,y)=0. \blacksquare

So, we can consider

\text{Bil}\left(\mathscr{U},\mathscr{V}\right)=\left\{B:\mathscr{U}\boxplus\mathscr{V}\to F\mid B\text{ is bilinear}\right\}

Consider though that if B_1,B_2\in\text{Bil}\left(\mathscr{U},\mathscr{V}\right) and \alpha\in F then if we define

B_3:\mathscr{U}\boxplus\mathscr{V}\to F:(x,y)\mapsto B_1(x,y)+B_2(x,y)

and

B_4:\mathscr{U}\boxplus\mathscr{V}\to F:(x,y)\mapsto \alpha B_1(x,y)

we can easily check that B_3,B_4\in\text{Bil}\left(\mathscr{U},\mathscr{V}\right). Thus, we may define B_1+B_2=B_3 and \alpha B_1 =B_4, and so noticing that

\bold{0}:\mathscr{U}\boxplus\mathscr{V}\to F:(x,y)\mapsto 0

is bilinear we can easily conclude that \text{Bil}\left(\mathscr{U},\mathscr{V}\right)\leqslant F^{\mathscr{U}\times\mathscr{V}} (where this latter space is the set of all mappings of the form \mathscr{U}\times\mathscr{V}\to F).

Basis and Dimension

With this insight, the first question is, if \dim_F \mathscr{U}=m and \dim_F\mathscr{V}=n what is \dim_F\text{Bil}\left(\mathscr{U},\mathscr{V}\right)? But, finding dimension is synonymous (for most finite-dimensional cases)  with finding a basis for the space. Thus, our real goal is to ascertain whether given a basis \{x_1,\cdots,x_m\} for \mathscr{U} and a basis \{y_1,\cdots,y_n\} for \mathscr{V} how we can create a related basis for \text{Bil}\left(\mathscr{U},\mathscr{V}\right) . The first step in this process is the following theorem:

Theorem: For any \left\{\gamma_{i,j}:i\in[m]\text{ and }j\in [n]\right\}\subseteq F there exits a unique B\in\text{Bil}\left(\mathscr{U},\mathscr{V}\right) such that B(x_i,y_j)=\gamma_{i,j}.

Proof: We first notice that if x\in\mathscr{U} and y\in\mathscr{V} that \displaystyle x=\sum_{i=1}^{m}\alpha_i x_i and \displaystyle y=\sum_{j=1}^{n}\beta_j y_j. Thus:

\displaystyle \begin{aligned}B(x,y) &= B\left(\sum_{i=1}^{m}\alpha_i x_i,\sum_{j=1}^{n}\beta_j y_j\right)\\ &= \sum_{i=1}^{m}\alpha_i B\left(x_i,\sum_{j=1}^{n}\beta_j y_j\right)\\ &=\sum_{i=1}^{m}\alpha_i\left(\sum_{j=1}^{n}\beta_j B(x_i,y_j\right)\\ &=\sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_i \beta_j B(x_i,y_j)\end{aligned}\quad\quad(1)

So, define

\displaystyle B(x,y)=B\left(\sum_{i=1}^{m}\alpha_i x_i,\sum_{j=1}^{n}\beta_j y_j\right)=\sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_i\beta_j \gamma_{i,j}

Evidently then B(x_i,y_j)=\gamma_{i,j} and so the existence is clear, and uniqueness follows by considering if B' were another such bilinear form then using (1) we may see that

\displaystyle B'\left(\sum_{i=1}^{m}\alpha_i x_i,\sum_{j=1}^{n}\beta_j y_j\right)=\sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_i \beta_j B(x_i,y_j)=\sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_i \beta_j \gamma_{i,j}=B\left(\sum_{i=1}^{m}\alpha_i x_i,\sum_{j=1}^{n}\beta_j y_j\right)

 

and so B=B'. \blacksquare

With this we are now able to produce an explicit basis for \text{Bil}\left(\mathscr{U},\mathscr{V}\right) given a basis for \mathscr{U} and one for \mathscr{V}.

Theorem: Let \{x_1,\cdots,x_m\} be a basis for \mathscr{U} and \{y_1,\cdots,y_n\} a basis for \mathscr{V}. Then,

\mathscr{B}=\left\{B_{i,j}:i\in[m]\text{ and }j\in[n]\right\}

is a basis where

B_{i,j}(x_p,y_q)=\delta_{i,p}\delta_{j,q}

Proof: The fact that such B_{i,j} exists is a consequence of the prior theorem. Suppose then that

\displaystyle \sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_{i,j}B_{i,j}=\bold{0}

Then, evaluating at (x_k,y_\ell), gives

\displaystyle 0=\sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_{i,j}B_{i,j}(x_k,y_{\ell})=\alpha_{k,\ell}

Thus, ranging k over \{1,\cdots,m\} and \ell over \{1,\cdots,n\} we find that

\alpha_{1,1}=\cdots=\alpha_{1,n}=\cdots=\alpha_{m,1}=\cdots=\alpha_{m,n}=0

and thus \mathscr{B} is linearly independent. To see that \text{span }\mathscr{B}=\text{Bil}\left(\mathscr{U},\mathscr{V}\right) we let B\in\text{Bil}\left(\mathscr{U},\mathscr{V}\right) be arbitrary. We claim that

\displaystyle B=\sum_{i=1}^{m}\sum_{j=1}^{n}B(x_i,y_j)B_{i,j}

To see this we note that if \displaystyle x=\sum_{i=1}^{m}\alpha_i x_i\in\mathscr{U} and \displaystyle y=\sum_{j=1}^{n}\beta_j y_j\in\mathscr{V} that

\displaystyle \begin{aligned}\sum_{i=1}^{m}\sum_{j=1}^{n}B(x_i,y_j)B_{i,j}(x,y)&=\sum_{i=1}^{m}\sum_{j=1}^{n}B(x_i,y_j)\left(\sum_{k=1}^{m}\sum_{\ell=1}^{n}\alpha_k\beta_{\ell} B_{i,j}(x_k,y_{\ell})\right)\\ &= \sum_{i=1}^{m}\sum_{j=1}^{n}\alpha_i\beta_j B(x_i,y_j)\\ &= B(x,y)\end{aligned}

Thus, \mathscr{B} is linearly independent and \text{span }\mathscr{B}=\text{Bil}\left(\mathscr{U},\mathscr{V}\right), thus \mathscr{B} is a basis for \text{Bil}\left(\mathscr{U},\mathscr{V}\right) as required. \blacksquare

From this we may gather the following rather aesthetically appealing corollary:

Corollary: \displaystyle \dim_F\text{Bil}\left(\mathscr{U},\mathscr{V}\right)=\dim_F\mathscr{U}\cdot\dim_F\mathscr{V}

References:

1. Halmos, Paul R. “Bilinear Forms.” Finite-dimensional Vector Spaces,. New York: Springer-Verlag, 1974. Print.

Advertisements

October 23, 2010 - Posted by | Algebra, Linear Algebra | , , ,

8 Comments »

  1. […] to Halmos’s section 23 on Bilinear forms, the content of which is explained (not quoted) here. Some interesting results in this post is the, very very perfunctory, discussion of quadratic forms […]

    Pingback by Halmos Section 23: Bilinear Forms « Abstract Nonsense | October 27, 2010 | Reply

  2. […] the  last post we discussed how given vector spaces over a field there is a canonical way to form the vector […]

    Pingback by Tensor Product « Abstract Nonsense | October 29, 2010 | Reply

  3. […] back when we discussed the concept of bilinear forms. Namely, for vector spaces over a vector space we decided that a bilinear […]

    Pingback by Multilinear Forms « Abstract Nonsense | November 11, 2010 | Reply

  4. […] my the last post we discussed how one can extend the notion of bilinear forms; one of the main reasons being because we desired to extend the notion of a determinant to higher […]

    Pingback by Symmetric and Antisymmetric forms « Abstract Nonsense | November 14, 2010 | Reply

  5. […] intuitively it seems as though the converse of the above is true since, for example taking bilinear forms we see that if is skew-symmetric […]

    Pingback by Alternating Forms « Abstract Nonsense | November 14, 2010 | Reply

  6. […] previously discussed the notion of bilinear forms as being maps which are linear in each entry (where are some […]

    Pingback by Representation Theory: The Tensor Product and the Tensor Product of Representations « Abstract Nonsense | February 14, 2011 | Reply

  7. […] could equivalently be thought of as declaring the canonical basis orthonormal and extending by bilinearity. Of course we can induce the usual norm on , , given by (as […]

    Pingback by Structure of Euclidean Space « Abstract Nonsense | May 22, 2011 | Reply

  8. […] recall that a bilinear form for some finite dimensional -space is one such that and are never the […]

    Pingback by Canonical Isomorphism Between a Finite Dimensional Inner Product Space and its Dual « Abstract Nonsense | June 4, 2011 | Reply


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

%d bloggers like this: