Changes in QMBF 5.10: --------------------- - added a new option for numerical differentiation (ND) in the Levenberg-Marquardt algorithm. For user-defined models, the derivatives of the model functions with respect to the fit parameters no longer need to be entered by the user when this option is active. For the predefined models, it is normally recommended to disable ND, because the built-in derivative implementations allow higher precision than ND. However, for the predefined models, ND provides a valuable test of the built-in derivative routines; the precision of ND is in fact completely sufficient in most cases. Indeed, the following bug was found by comparing the results to those from ND: - fixed a bug in the calculation of derivatives of the model functions with respect to the "oo" amplitude parameters in the following models: "3-Point Correlator, Multiple (incl. oscillating) Exponentials" "3-Point Correlator, Multiple (incl. oscillating) Exponentials with Exponential Energies" "3-Point Correlator, Vector Fit, Multiple (incl. oscillating) Exponentials" "3-Point Correlator, Vector Fit, Multiple (incl. oscillating) Exponentials with Exponential Energies" Detailed description of change: line 242 of "threept_multi_alt_exp_model.cpp" and "threept_multi_alt_exp_expE_model.cpp" has been corrected from previously return alt(t)*alt(T-t)*eo_model->eval_derivative(function, mexp_final+mexp_initial+parameter, arguments); to return alt(t)*alt(T-t)*oo_model->eval_derivative(function, mexp_final+mexp_initial+parameter, arguments);