Hi,
I am programming a maximum likelihood estimator in Mata, and have programmed a d2 evaluator that returns the log-likelihood, gradient, and Hessian. This works very well with the Newton-Raphson algorithm, but I would like to provide the functionality for the user to select the optimization technique they want. After trying using the same evaluator both as d1 and d0 and getting errors, I am wondering if for any other optimization technique, we have to use a gf2 estimator, so that you return all the scores, not just the sum of them. Honestly, I am just out of ideas of why I can't use the same evaluator to estimate the model with either the DFP or BFGS method instead of NR. But when I do, the log-likelihood goes to 0 and it gives errors.
