manopt solver fiailed

[jifa zhang]

hello, when i use manopt solver to deal with my problem, something wrong, can you help me? thank you very much.

[Ronny Bergmann]

Hi there,

can you maybe provide a bit more details, a bit of code that reproduces the problem, that we can run as well, maybe?

This way I can just tell you that for the silver you used (it seems to be trust regions?) gets a rho that is NaN, which – as the error message also says – might for example be due to the retraction returning a point outside of the manifold. This can be numerical inaccuracies or that you are on a manifold where the retraction you use only makes sense locally.

[Nicolas Boumal]

The cost function value seems to be NaN already for the initial guess (line "iter 0" shows "Cost val" to be NaN). This suggests an issue with the cost function implementation.