Alternative parametrizations of linear transformations?

34 Views Asked by At

Matrices seem to be the most common parametrization of linear transformations with the geometrical meaning of columns of the matrix being the re-mapped basis vectors. But sometimes it's difficult to visualize what exact effect will a linear transform (represented by a matrix) have on, let's say, the unit circle.

For example, I see eigenvectors and eigenvalues as an alternative parametrization of linear transformations, because they tell in which directions and by how much does a linear transformation stretch space, although some linear transformations have complex eigenvectors and eigenvalues, which breaks this stretching analogy for me. Another flaw in this analogy for me is that I fail to see why e.g. $2\times 2$ matrices cannot encode more than 2 axes of stretching (more than 2 eigenvectors).

When playing with 2D basis vectors, I saw that shear transforms produce results similar to camera movement in 3D space. Maybe N-dimensional transforms can be also expressed as certain operations on (N+1)-dimensional space?

Are there any other useful parametrizations of linear transformations that could perhaps bring more intuitive insight into how non-trivial (non-orthogonal) linear transformations affect space?