Dear Frédéric,
Thank you for your answer. Upon your advice, I have found a bypass, by using the fact that my space is a subspace of a known fespace on the split mesh Ths=splitmesh6PowellSabin(Th0).
I have performed a “space reduction” similar to what is done with on(tgv=-1): I replaced some well chosen lines in my matrix by constraints defining the subspace.
Nevertheless I would be interested in learning to define fespaces, and understanding the meaning of all the parameters involved. Probably other users would be interested also (for example What does the array of ``data[ ]'' represent during the process of adding a new finite element? ). I don’t think there is a tuto on that.