On local identifiability of linear systems (Q1057198)

This paper states new necessary and sufficient conditions for local identifiability; a condition for the model to be locally identifiable almost everywhere and a condition which permits drawing conclusions for each specific point in the parametric space. These criteria are of very general significance, easily programmable, and readily applied, as demonstrated by many given examples. The author considers the model described by the following linear differential system: \[ dN(t)/dt=AN(t)+Bu(t),\quad t\in [0,T],\quad N(0)=0,\quad y(t)=CN(t), \] where N is a vector function with dimension n, A is an \(n\times n\) square matrix, B is a matrix with dimensions \(n\times b\) and C is a matrix with dimensions \(c\times n\). The main results are the following: Theorem 1: A sufficient condition for local identifiability is that the determinant of K'K is different from zero, where the matrix K is suitably defined. Theorem 2: A necessary and sufficient condition for a model to be locally identifiable almost everywhere is that the determinant of K'K is different from zero almost everywhere. Theorem 3: A necessary and sufficient condition for local identifiability of a solution P is the existence of a stage k at which the determinant of \(K^{(k)t}K^{(k)}\) is different from zero \((K^ t=K')\).