Matrices have a least two major functions in linear algebra. On one hand, they can represent linear transformations as elements of $\text{Hom}(V,V)$). On the other hand they can represent inner products as elements of $\text{Hom}(V,V^*).$

In the case of linear transformations, there is a clear geometrical interpretation of eigenvalues and eigenvectors - eigenvectors are scaled by eigenvalues. For a matrix representing an inner product, it's not as clear to me. I don't expect the eigenvalues of an inner product matrix to be intrinsic to the inner product since eigenvalues of an inner product matrix are not preserved by a change of basis. Is there however a way to interpret the eigenvalues in relation to a chosen basis?

This seems to be a reasonable hope since there are relationships between an inner product matrix and its eigenvalues. For example, a symmetric inner product matrix (say over $\mathbb{R}$) with all positive eigenvalues represents a positive definite inner product.

  • $\begingroup$ observation: if $B$ is the matrix got by the change of basis due to the eigenvectors of $Q$, where $Q$ is the matrix of a inner product, then the product $B^{\top}QB$ will be the reinterpretation of the very same inner product given by $Q$ but in the new basis $\endgroup$ – janmarqz Jul 15 '15 at 16:40

The matrix $A$ which induces an inner product must be symmetric (or hermitian), non-degenerate and positive definite.

Because of its symmetry, we have using the spectral theorem, a full set of eigenvalues and corresponding eigenvectors and therefore $A$ is diagonalizable with positiv eigenvalues. Even under the change of the basis, $A$ has still the same eigenvalues since matrix similarity states that similar matrices have the same eigenvalues.

A matrix $A$ is similar to $B$ if there exists invertible $S$ such that $$ A=S^{-1}BS $$


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.