joachim wrote: Your observation is correct:
If you apply the source at the symmetry plane of the full model, the weak form leads to the interface condition
[du/dn] = g
i.e.
du/dn_left + du/dn_right = g,
i.e. by symmetry
du/dn_left = du/dn_right = g/2
while for the half problem the source term is
du/dn = g
in physical words:
if you apply the source term for the full problem, half of the energy is going to the left, and half to the right.
if you do the half problem, the full energy is going to this side.
s
Joachim
I am not sure if I understand your explanation: are you effectively saying then only half the source magnitude must be applied when doing load integrals in case of half problem? Perhaps it would work for a point source but not for distributed loading on the surface. In this case, either for a full or a half problem, the Neumann data g is integrated on a specific area. Wouldn't ngsolve do the integral contributing to the load vector f as:
[tex]
f_i = \int_{\Gamma_1} g \phi_i ds
[/tex]
where \phi_i are the basis functions. So integral for the full problem will be performed on a full circle whereas that for the half problem is performed only on the half. So although the magnitude of the source term is the same in both cases, the half model only "sees" half its contribution due to the affected area being half. Thus no scaling should be required when dealing with half model (again assuming that the dp/dn=0 exists in the original problem and nothing needs to be said on the symmetrical plane) Does it make sense?
