This module is for dealing with quantities of interest within the framework of verification, validation, and uncertainty quantification.
method
__init__
keywords:
norm_factor
When coupling the minimization of error between a prediction x*
and the target value x
, it is helpful that the error terms are of approximately the same size so that minimization of coupled objective functions aren't dominated by a larger error which is due a a larger target value. For example, the cohesive energy of structure are measured in the range of 2eVeV/atom to 6eV/atom, lattice parameters are measured in the range 5 Angstroms, while elastic tensors are measure in the 100-500 GPa range. Coupling objective functions using a technique such as weighted sum of squares
C = w['e_coh']qoi['e_coh'].sse + w['a_latt']qoi['a_latt'].sse + w['B_modulus'] qoi['B_modulus'].sse
would be completely dominated by the third term.
However, if we div