PiNNs solving PDEs fail to converge to right solutions

1.why would PiNNs solving PDEs converge to wrong solutions? That is when the loss seems to converge to 0, the solution is far from the analytic.
2.why would PiNNs solving PDEs jump out a good solution towards a wrong one? That is the loss first seems to converge to 0 while the solution looks good but then increases accompanied by the solution becoming worse.

is there any answer?