A procedure for generating locally identifiable reparametrisations of unidentifiable nonlinear systems by the similarity transformation approach (Q1264487)

One shows first that the maximum number of locally identifiable parameters by which the system may be reparametrized is given by the rank of a certain Jacobian matrix related to the functions defining the given system. Then the reparametrization may be generated by an orthogonality condition and by solving the corresponding partial differential equations in order to find locally identifiable combinations of original parameters and finally to establish a (non unique) locally identifiable reparametrisation of the given system. The reduction of the parameter number is equal to the rank deficiency of the Jacobian matrix and the most appropriate reparametrisation has also to be meaningful relative to the given physical system. Two examples illustrate the usefulness and some drawbacks of the procedure.