about ManoptAD

[Fengcheng Pei]

Hello,

In my program, there is no random number. However, when I repeat running it, the optimization result is different and sometimes it said that “Automatic differentiation for gradient failed”. I am very confused, because it is exactly the same program without any random number. Whether I must write down the formula of gradient then I will generate the same result?

Besides, I also want to know a warning “Gradient and Hessian already defined”. Does this warning matter?

Thanks for your help and hope for response!

Best regards,

Fengcheng Pei

[Nicolas Boumal]

Hello,

The warning  “Gradient and Hessian already defined” is important: it suggests that the problem structure may contain several competing (and perhaps incompatible) definitions of the gradient and the Hessian.

Between several runs, make sure to clear the problem structure, e.g., with "clear all;".

For your other questions, it is difficult to know without seeing more of your code. Can you share the most important bits here?

Nicolas