4
$\begingroup$

In Linear Algebra, when computing an inner product $<x,y> = y^*Wx$, what is the name of the matrix W?

If it doesn't have a name, where can I find a practical explanation of how to construct it for a particular problem or space?

Is there a text on the subject that explains this simply? A note on my background, I am approaching this from the perspective of an engineering student and not that of a mathematician; I don't have an understanding of the finer points of topology or differential geometry (not yet at least :) ).

I believe that this is the same matrix used in vector calculus to perform a change of variables. As in, we perform our change of variables, then multiply the new expression by $\frac{det(J(W))}{det(J(V))}$ where $W$ is this magic matrix in the new space, $V$ is this magic matrix the old space, and J(X) is the Jacobian operator. Am I correct?

Also, is this related to one of the the matrices that comes up in Singular Value Decomposition (namely the diagonal matrix containing the singular values)?

$\endgroup$
6
  • 3
    $\begingroup$ I would just call W "the matrix of the inner product." If you get W from the standard inner product by a change of variables, then it will be a Gram matrix: en.wikipedia.org/wiki/Gramian_matrix $\endgroup$ Nov 29, 2010 at 9:19
  • $\begingroup$ How do I form this matrix for a particular problem? I realize that I need to 'define' orthogonality through it; how do I do that? $\endgroup$
    – hlzr
    Nov 29, 2010 at 10:09
  • 1
    $\begingroup$ What kind of particular problem? $\endgroup$ Nov 29, 2010 at 10:34
  • $\begingroup$ SVD is a different beastie; briefly, it's a numerically stable way of simultaneously determining the eigensystems of $\mathbf A\mathbf A^T$ and $\mathbf A^T\mathbf A$ when you are given $\mathbf A$. The singular values of $\mathbf A$ are the eigenvalues of those matrix products. $\endgroup$ Nov 29, 2010 at 12:54
  • $\begingroup$ For constructing an appropriate inner product, it's application-dependent, as Qiaochu says. $\endgroup$ Nov 29, 2010 at 12:55

1 Answer 1

3
$\begingroup$

I think you will need to give more details on what is motivating your search for $W$ before you'll get a satisfactory answer. I will make a guess that you have a vector space that has a preferred basis and you are seeking an inner product which renders this basis orthonormal.

I don't know of a name for $W$; it is the matrix of the inner product. It should be (symmetric) positive definite.

Suppose we have $d$ linearly independent vectors $\{x_i\}$ in $\mathbb{R}^d$ that are a convenient basis for our work, except that they aren't orthonormal. We may as well assume they are normalized with respect to the usual inner product on $\mathbb{R}^d$. Make a matrix $X$ whose columns are these vectors. Then $X$ is the matrix of a change of basis. Specifically, $X^{-1}y=x$ gives us $x$ as the representation of $y$ in the basis $\{x_i\}$.

Now consider $x^Tx$. That is the inner product we want. It is the inner product which renders our basis orthonormal. Since $x^Tx = y^TX^{-T}X^{-1}y$, we can set $W = X^{-T}X^{-1}$ and we have the desired inner product without having to explicitly perform a change of basis. (Here $X^{-T}$ is the inverse of the transpose of $X$.)

One context where I have dealt with this is when $X$ was the complete basis of eigenvectors of a linear operator, and I wanted an inner product for which the operator was self-adjoint.

$\endgroup$
3
  • $\begingroup$ Thanks so much! That is really quite a clear explanation. $\endgroup$
    – hlzr
    Nov 29, 2010 at 18:17
  • $\begingroup$ Also, I don't have enough rep to upvote your answer, but I can say thank you :) $\endgroup$
    – hlzr
    Nov 29, 2010 at 19:00
  • $\begingroup$ Thanks, and you are welcome. $\endgroup$
    – yasmar
    Nov 29, 2010 at 20:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.