What is the difference between SVD and SVD-Assist?
When PEST uses truncated singular value decomposition (SVD) to solve the inverse problem of model calibration, it decomposes parameter space into two orthogonal subspaces spanned by orthogonal unit vectors. The unit vectors which span the calibration solution space represent combinations of parameters that are estimable on the basis of the current calibration dataset. Those spanning the null space comprise inestimable parameter combinations. Using standard techniques, PEST only estimates those parameter combinations which are in fact estimable, leaving the values of inestimable parameter combinations unchanged. (Actually, PEST works in terms of parameter perturbations from initial user-supplied parameter values rather than absolute parameter values – but the principle is the same.) Decomposition of parameter space in this manner takes place on the basis of the Jacobian matrix. This is the matrix of sensitivities of all model outputs corresponding to observations comprising the calibrat
