Remove printing that's specific to gradient-based VQE #262
Conversation
| "--------------------------------------------------------------------------------------------------" | ||
| ) | ||
| header = "\n k iteration Energy dE" | ||
| if self._use_analytic_grad: |
There was a problem hiding this comment.
I have a suggestion for future consideration. The decision to print gradient diagnostics should ideally depend on the type of optimizer being used, whether it is gradient-based or gradient-free. Even when numerical gradients are employed, gradient diagnostics should still be printed. However, scipy.optimize.minimize does not directly expose numerical gradients to the callback function. Therefore, the current decision to report gradient diagnostics for numerical gradients is based on self._use_analytic_grad.
| "--------------------------------------------------------------------------------------------------" | ||
| ) | ||
| header = "\n k iteration Energy dE" | ||
| if self._use_analytic_grad: |
There was a problem hiding this comment.
Looking at uccnvqe.py, it appears that a user could potentially request optimizer = "Nelder-Mead" and use_analytic_grad = True. More or less the same applies to adaptvqe.py. To ensure accurate printing of diagnostics, I suggest internally setting self._use_analytic_grad to False if a gradient-free optimizer is selected. Additionally, a warning message could be printed to inform the user of this change.
Description
Previously, UCCVQE would print out gradient diagnostics even for computations without a gradient. This is silly, so I removed them.
To clean up the printing, I also tried to unify the to-screen printing and to-file printing. For some reason I don't understand, to-file printing had more digits but not more space for them. There were also some inconsistencies with the amount of whitespace between the two versions. In both cases, I changed them to be more like the to-screen version. The initial
#for to-file printing is retained.User Notes
Checklist