I recently added new functionality. This is a reminder for myself; still have to document:

Optimizer properties (counters):

• optimizer.iter (The number of the current iteration. Starts from optimizer.init_iter)
• optimizer.iter_count (How many iterations have been executed)
• optimizer.function_eval_count (How many times the objective function has been called)
• optimizer.fidelity_eval_count (How many times the fidelity function has been called)

Other functionality:

• optimizer.make_params_collector() which returns an empty list, that will be populated with the new parameters each time the optimizer performs an iteration.
• optimizer.restart() to reset all counters.
• Second-order optimization does not require anything else. When defining the optimizer, you must declare second_order=True, and the rest is handled internally.
• Hessian postprocessing methods for second_order or quantum_natural.
• How to define and pass the fidelity function for quantum_natural optimization.
• What is the scalar hessian approximation.

Changed features:

• By default a callback should return nothing. If anything is returned, it will be considered as a flag to stop iterating. You can use it to check convergence and stop iterating if achieved.
• The optimizer now does not have a maximum number of iterations. To specify how many times you want to iterate, use optimizer.run(fun, guess, num_iter=...).
• Now, instead of using postprocessing, you can define optimizers with the option apply_update(guess, update) which returns the guess for the next iteration. By default, apply_update = np.add, and therefore, new_guess = np.add(guess, update).