covariance

Determine how the final covariance matrix is computed

Specification

  • Alias: None

  • Arguments: INTEGER

  • Default: 0 (no covariance)

Description

covariance (NL2SOL’s covreq) specifies whether and how NL2SOL computes a final covariance matrix.

The desired covariance approximation:

  • 0 = default = none

  • 1 or -1 ==> \(\sigma^2 H^{-1} J^T J H^{-1}\)

  • 2 or -2 ==> \(\sigma^2 H^{-1}\)

  • 3 or -3 ==> \(\sigma^2 (J^T J)^{-1}\)

  • Negative values ==> estimate the final Hessian H by finite differences of function values only (using fd_hessian_step_size)

  • Positive values ==> differences of gradients (using fd_hessian_step_size)