"Mathematics is the art of giving the same name to different things."-Henri Poincare 3.1 Introduction Any square matrix A of size d × d can be considered a linear operator, which maps the d-dimensional column vector x to the d-dimensional vector Ax. A linear transformation Ax is a combination of operations such as rotations, reflections, and scalings of a vector x. A diagonalizable matrix is a special type of linear operator that only corresponds to a simultaneous scaling along d different directions. These d different directions are referred to as eigenvectors and the d scale factors are referred to as eigenvalues. All such matrices can be decomposed using an invertible d × d matrix V and a diagonal d × d matrix Δ: A = V ΔV −1 The columns of V contain d eigenvectors and the diagonal entries of Δ contain the eigen-values. For any x ∈ R d , one can geometrically interpret A x using the decomposition in terms of a sequence of three transformations: (i) Multiplication of x with V −1 computes the coordinates of x in a (possibly non-orthogonal) basis system corresponding to the columns (eigenvectors) of V , (ii) multiplication of V −1 x with Δ to create ΔV −1 x dilates these coordinates with scale factors in Δ in the eigenvector directions, and (iii) final multiplication with V to create V ΔV −1 x transforms the coordinates back to the original basis system (i.e., the standard basis). The overall result is an anisotropic scaling in d eigenvector directions. Linear transformations that can be represented in this way correspond to diagonalizable matrices. A d × d diagonalizable matrix represents a linear transformation corresponding to anisotropic scaling in d linearly independent directions. When the columns of matrix V are orthonormal vectors, we have V −1 = V T. In such a case, the scaling is done along mutually orthogonal directions, and the matrix A is always © Springer Nature Switzerland AG 2020 C. C. Aggarwal, Linear Algebra and Optimization for Machine Learning, https://doi.
CITATION STYLE
Aggarwal, C. C. (2020). Eigenvectors and Diagonalizable Matrices. In Linear Algebra and Optimization for Machine Learning (pp. 97–139). Springer International Publishing. https://doi.org/10.1007/978-3-030-40344-7_3
Mendeley helps you to discover research relevant for your work.