Many of the most effective function optimization algorithms require the gradient of the function to be optimized. In many cases of practical interest the gradient is not available in closed form and must be approximated numerically, finite differences being the most frequent approach. This note suggests a number of reasonableness tests which can be performed on the gradient vector to aid the user in identifying the erroneous problem formulations and poor problem scaling.
/lp/association-for-computing-machinery/reasonableness-tests-for-gradient-based-function-minimization-w0dWd3cOaJ