Abstract Nonsense

Crushing one theorem at a time

Symmetric and Skew-Symmetric forms

Point of post: In this post I discuss the equivalent of section 29 and the end of 30, in more detail than is done in Halmos’s book.


In 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 dimensions (in some intuitive sense). It turns out though that the generality to which we extended it is often (not always) too general for the level of linear algebra we are operating at. In particular, instead of studying \text{Mult}\left(\mathscr{V}_1,\cdots,\mathscr{V}_n\right) when \mathscr{V}_1,\cdots,\mathscr{V}_n are any arbitrary, possibly distinct vector spaces over some field F we instead intend to study \text{Mult}\left(\mathscr{V}_1,\cdots,\mathscr{V}_n\right) where \mathscr{V}_1=\cdots=\mathscr{V}_n=\mathscr{V}. In this case we denote \text{Mult}\left(\mathscr{V}_1,\cdots,\mathscr{V}_n\right) by \text{Mult}_n\left(\mathscr{V}\right). The reason why we would want to consider this scenario is because when we restrict our attention to multilinear forms on the direct sum of the same vector space the resultant space has some beautiful interplay with S_n. In other words,  there is a very natural way in which S_n interacts with the elements of \text{Mult}_n\left(\mathscr{V}\right). Our first indication of this was symmetric bilinear forms. Which, phrased slightly differently say that B(x_1,x_2)=B(x_{\pi(1)},x_{\pi(2)}) for every \pi\in S_2, which isn’t very profound considering there are only two elements of S_n. But, we will be able to go further and say, in a precise manner yet to be said, that S_n acts on \text{Mult}_n\left(\mathscr{V}\right) so that there is a meaningful way to “multiply” elements of S_n and elements of \text{Mult}_n\left(\mathscr{V}\right). This, is where all our work with permutations starts to come into play (though, to be quite honest I probably developed the theory a little further than it really needed to be).

Remark: Halmos defines something in his book either incorrectly, or ascribes to it a property which isn’t true (see below).

S_n‘s Action On \text{Mult}_n\left(\mathscr{V}\right)

Let \mathscr{V} be a finite dimensional vector space over F. Then, for every \pi\in S_n and every K\in\text{Mult}_n\left(\mathscr{V}\right) there is a way to associate the two to create a new element \pi K of \text{Mult}_n\left(\mathscr{V}\right) given by (\pi \cdot K)\left(x_1,\cdots,x_n\right)=K\left(x_{\pi^{-1}(1)},\cdots,x_{\pi^{-1}(n)}\right). For example, if

\pi=\begin{pmatrix}1 & 2 & 3 & 4\\4 & 2 & 1 & 3\end{pmatrix}

then (\pi \cdot K)(1,3,9,2)=K(2,3,1,9). Indeed this is a n-form since

\begin{aligned}\left(\pi\cdot K\right)\left(x_1,\cdots,\alpha x_k+\beta x'_k,\cdots,x_n\right) &= K\left(x_{\pi^{-1}(1)},\cdots,\alpha x_{\pi^{-1}(k)}+\beta x'_{\pi^{-1}(k)},\cdots,x_{\pi^{-1}(n)}\right)\\ &= \alpha K\left(x_{\pi^{-1}(1)},\cdots,x_{\pi^{-1}(k)},\cdots,x_{\pi^{-1}(n)}\right)+\beta K\left(x_{\pi^{-1}(1)},\cdots,x'_{\pi^{-1}(k)},\cdots,x_{\pi^{-1}(n)}\right)\\ &=\alpha\left(\pi\cdot K\right)\left(x_1,\cdots,x_k,\cdots,x_n\right)+\beta\left(\pi\cdot K\right)\left(x_1,\cdots,x'_k,\cdots,x_n\right)\end{aligned}

So, with this we can state the main reason for restricting our attention to \text{Mult}_n\left(\mathscr{V}\right):

Theorem: Let \mathscr{V} be a finite dimensional F-space, then S_n acts on \text{Mult}_n\left(\mathscr{V}\right) via the action

\cdot:S_n\times\text{Mult}_n\left(\mathscr{V}\right)\to\text{Mult}_n\left(\mathscr{V}\right):\left(\pi,K\right)\mapsto \pi\cdot K


\displaystyle \left(\pi\cdot K\right)(x_1,\cdots,x_n)=K\left(x_{\pi^{-1}(1)},\cdots,x_{\pi^{-1}(n)}\right)

Proof: It suffices to prove that \text{id}_{[n]}\times K=K for all K\in\text{Mult}_n\left(\mathscr{V}\right) and

\sigma\cdot\left(\tau\cdot K\right)=\left(\sigma\tau\right)\cdot K

for all \sigma,\tau\in S_n. So, the first of these is clear since

\left(\text{id}_{[n]}\cdot K\right)\left(x_1,\cdots,x_n\right)=K\left(x_{\text{id}^{-1}{[n]}},\cdots,x_{\text{id}^{-1}_{[n]}(n)}\right)=K(x_1,\cdots,x_n)

for all x_1,\cdots,x_n\in\mathscr{V}.  For the second condition we merely note that

\begin{aligned}\left(\sigma\cdot\left(\pi\cdot K\right)\right) &=\left(\pi\cdot K\right)\left(x_{\sigma^{-1}(1)},\cdots,x_{\sigma^{-1}(n)}\right)\\ &= K\left(x_{\pi^{-1}\left(\sigma^{-1}\left(1\right)\right)},\cdots,x_{\pi^{-1}\left(\sigma^{-1}\left(n\right)\right)}\right)\\ &= K\left(x_{(\sigma\pi)^{-1}(1)},\cdots,x_{(\sigma\pi)^{-1}(n)}\right)\\ &= \left(\left(\sigma\pi\right)\cdot K\right)\left(x_1,\cdots,x_n\right)\end{aligned}

and since x_1,\cdots,x_n\in \mathscr{V} were arbitrary it follows that \sigma\cdot\left(\pi\cdot K\right)=\left(\sigma\pi\right)\cdot K as required.  \blacksquare

From hereon out we drop the \cdot in \pi\cdot K in lieu of \pi K.

We are now prepared to discuss two important classes of n-forms whose definitions are precipitated by the above theorem. Namely the generalization of symmetric bilinear forms t and their counterparts skew-symmetric bilinear forms. More explicitly, we call K\in\text{Mult}_n\left(\mathscr{V}\right) symmetric if \pi K=K for all \pi\in S_n (equivalently that the isotropy subgroup generated by K is the full group, for those  who know the jargon) and we call K skew-symmetric if \pi K=\text{sgn}(\pi)K for all \pi\in S_n.

Now, with this it’s clear that not all n-forms are symmetric, nor are all skew-symmetric, in fact not all are either. That said, we will see that from each n-linear form K there is a natural way for one to build a symmetric and an skew-symmetric form. Namely:

Theorem: Let \mathscr{V} be a finite dimensional F-space and K\in\text{Mult}_n\left(\mathscr{V}\right) then the n-form \text{Sym}\left(K\right) given by

\displaystyle \text{Sym}\left(K\right)=\sum_{\pi\in S_n}\pi K

is symmetric.

Proof: Let \sigma\in S_n be fixed but arbitrary and let \sigma S_n=\left\{\sigma \pi:\pi\in S_n\right\}. We claim that \sigma S_n=S_n. To do this it suffices to show that

f_\sigma:S_n\to S_n:\pi\mapsto \sigma\pi

is a bijection. To prove injectivity we merely recall that \sigma\pi=\sigma\pi'\implies \pi=\pi' and surjectivity follows from f_{\sigma}\left(\sigma^{-1}\pi\right)=\pi for all \pi\in S_n from where bijectivity follows( this is really just a statement of Cayley’s theorem for S_n). Thus, we see that

\displaystyle \sigma\text{Sym}\left(K\right)=\sigma\sum_{\pi\in S_n}\pi K=\sum_{\pi\in S_n}\sigma\pi K=\sum_{\tau\in \sigma S_n}\tau K=\sum_{\tau\in S_n}\tau K=\text{Sym}\left(K\right)

from where it follows by the arbitrariness of \sigma that \text{Sym}\left(K\right) is symmetric. \blacksquare

As would be expected there is a analogous way to naturally procure an skew-symmetric form from any n-form K, namely:

Theorem: Let \mathscr{V} be a finite dimensional F-space and K\in\text{Mult}_n\left(\mathscr{V}\right). Then, the n-form \text{Asym}\left(K\right) given by

\displaystyle \text{Asym}\left(K\right)=\sum_{\pi\in S_n}\text{sgn}(\pi)\pi K

is skew-symmetric.

Proof: Let \sigma\in S_n be arbitrary. Using the fact that \sigma S_n=S_n as proven in the last theorem we see that

\displaystyle \begin{aligned}\sigma\text{Asym}\left(K\right) &=\sigma\sum_{\pi\in S_n}\text{sgn}(\pi)\pi K\\ &=\sum_{\pi\in S_n}\frac{1}{\text{sgn}(\sigma)}\text{sgn}\left(\sigma\pi\right)\sigma\pi K\\ &=\text{sgn}(\sigma)\sum_{\tau \in \sigma S_n}\text{sgn}(\tau)\tau K\\ &=\text{sgn}(\sigma)\sum_{\tau\in S_n}\text{sgn}(\tau)\tau K\\ &=\text{sgn}(\sigma)\text{Asym}\left(K\right)\end{aligned}

where we’ve made tacit use of the fact that \text{sgn}(\sigma\pi)=\text{sgn}(\sigma)\text{sgn}(\pi) (proof can be be found here) and that \displaystyle \frac{1}{\text{sgn}(\sigma)}=\text{sgn}(\sigma) (since \text{sgn}(\sigma)=\pm 1). Thus, since \sigma was arbitrary the conclusion follows. \blacksquare

Remark: \text{Asym} is for antisymmetric another name for skew-symmetric forms.


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


November 14, 2010 - Posted by | Algebra, Halmos, Linear Algebra, Uncategorized | , , , ,


  1. […] in our last post we restricted our study of multilinear algebra to that of spaces of the form in the attempt to […]

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

  2. […] let where is the “antisymmetrizer function” described in the post on skew-symmetric forms. Note then that if then (note that this no longer evaluates to zero on […]

    Pingback by Halmos Section 29,30 and 31 « Abstract Nonsense | November 18, 2010 | Reply

  3. […] Remark: If the notation seems unfamiliar, that’s because I created my notation. See here and here […]

    Pingback by Halmos Sections 32 and 33: Linear Transformations and Transformations as Vectors (Pt. I) « Abstract Nonsense | November 22, 2010 | 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 )

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s

%d bloggers like this: