I estimate survival models. However, the model converges sometimes and sometimes it doesn't even though it is exactly the same line that I run.
Every other time the output says "Warning: variance matrix is nonsymmetric or highly singular" and while the coefficients are the same there are no standards errors etc. (only dots).

This is true for Cox regressions with stcox and also with parametric regressions with streg.

Does anyone here know why this happens?

Thank you.