Cholesky Decomposition with Non-PD Fallback — .hurdle_chol_or

Internal helper shared by the RE-transform and marginal-prediction code paths. Attempts chol(Sigma); if Sigma is not positive definite (chol() errors — near-boundary rhos, overflow, or tanh(raw) rounding to exactly +/-1), falls back to the Cholesky factor of an uncorrelated diagonal covariance built from sigma_diag and emits ONE classed warning, since the returned/reported correlation estimates no longer describe the transformed random effects or marginal draws that result (TICKET-061).

Usage

.hurdle_chol_or_fallback(Sigma, sigma_diag)

Arguments

Sigma: A candidate covariance matrix.
sigma_diag: Numeric vector of per-RE variances for the diagonal fallback (length must equal nrow(Sigma)).

Value

The upper-triangular Cholesky factor of Sigma, or of the diagonal fallback when Sigma is not positive definite.