Dear Yves,
Thank you very much. Now it works perfectly. It was indeed the problem of dimension of element_K which was at the origin of the error. I have downloaded the fixed version since the use of ":" did not work because of mismatch between dimensions (2x2 : 2x1).
The _expression_ "Grad_u:Grad_Test_u" is not equivalent concerning the contribution of the longitudinal part of the work since it also includes a part of work of the orthogonal displacement. The latter does not contribute to the extension work at first order but it rather contributes to the term related to the flexion expressed by means of its second derivative obtained by the Hessian. I've also built it and it works well (the analytical solution of the simple bending beam is retrieved).
I have been initially misled by the fact that the gradient of a vector for a 1D structure is actually a 2x2 in 2D (or 3x3 in 3D) matrix equal to du/address@hidden where t is the vector "Normalized(element_K)" in GetFEM instead of just the vector du/ds as I would have expected since s is the only local coordinate. And the hessian is the third order tensor d2u/address@hidden@t instead of the vector d2u/ds2. But these vectors du/ds and d2u/ds2 can be obtained by contracting Grad_u with "Normalized(element_K)" on the one hand and Hess_u with "Normalized(element_K)@Normalized(element_K)" on the other hand. Once these operations are performed, everything is OK now and GetFEM works perfectly for arbitrary curvilinear structures.
Thank you again.
Best regards
Jean-François