The design matrix during GLS fitting is a concatenation of the timing model design matrix and the noise model design matrix. The noise designmatrix does not change during the fit and so the noise sub-block should be able to be completely cached for the duration of the entire fit. A similar thing is true for the entire covariance matrix since the timing model components of the covariance matrix are typically assumed to be improper.
This is especially true but less obvious to me in get_gls_mtcm_mtcy()
The design matrix during GLS fitting is a concatenation of the timing model design matrix and the noise model design matrix. The noise designmatrix does not change during the fit and so the noise sub-block should be able to be completely cached for the duration of the entire fit. A similar thing is true for the entire covariance matrix since the timing model components of the covariance matrix are typically assumed to be improper.
This is especially true but less obvious to me in
get_gls_mtcm_mtcy()