Skip to content

Arbitrary matrix function, or one of the listed ones  #2

Description

@peekxc

For $A \in \mathbb{R}^{n \times n}$ admitting an eigenvalue decomposition $A = U \Lambda U^T$ and $f : \mathbb{R} \to \mathbb{R}$ is a scalar function, according to the docs this package helps to estimate the quantity:

$$ \mathrm{tr}(f(A)) = \mathrm{tr}(U f(\Lambda) U^\top)$$

A couple of quantities I'm interested in computing:

  1. The numerical rank of $A$ (or the eigencount in an interval), per one of the applications
  2. The spectral density (also one of the applications)
  3. The trace of an arbitrary matrix function, e.g. functions like $\lambda \mapsto \lambda / (\lambda + \epsilon)$ (often used in $\ell_1$/rank minimization tasks) or $\lambda \mapsto \exp(-t \lambda)$ (related to the heat kernel)

Though some of these are mentioned, the only one's I see in the API docs are logdet, trace, traceinv, and schatten. Does this package support (1) and (2), and does it support arbitrary matrix functions like (3) [if not from Python, what about on the C++ side?]?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions