“…If it is not practical to calculate all N S values of
, due to computational expense, a subset N U of the overall set of post burnβin models may be used (see Section 4). The relationship between the uncertainty on a physical prediction,
, and the uncertainty on the underlying model parameters,
, is dependent on the sensitivity of
to each parameter,
(i.e., the gradient,
), and the covariance structure of the model,
(Champac & Garcia Gervacio,
2018). In the case of the anelasticity parameterization,
and
are nonβlinear functions of V S , complicating the analytical calculation of their expectation value and variance.…”