Thank you very much for your answer!
So indeed the example I was mentioning is here:
The main differences are related to the fact that they are using the symmetric gradient - which for now we left it simple -, the use of an exact solution to compare with, and finally a restriction of the domain.
I’m wondering if it’s related to the latest the problem that we encouter and to refer to what you were suggesting about the 0 problem. In their case, there should be only a 0 axis on the border, whereas in our case it’s in the middle of the domain …