> Upon closer inspection, one can see that the bottom left corner looks a bit
> "shaky" on the first 2 figures.
This is an artifact of plotting. What you are trying to visualize is a
bilinear function on the bottom left cell, but Paraview/Visit actually show is
a subdivision of the square cell into two triangles, and then using a linear
interpolation on each triangle. The way the square is subdivided into
triangles uses the bottom right -> top left diagonal. As a consequence, the
bottom left and the bottom right cell are not shown the same way, even their
values are perfectly symmetric.
> I tried to get to the bottom of why this happens numerically ?
> I found that at some point during my assembly of the RHS, there is a
> difference in what I obtained even though it should be perfectly symmetric.
Are you observing this visually, or are you looking at actual numbers?
> By digging even more I realised that the discrepancy steams from the
> boundary_worker (similar to the one of step-12).
> On the fig. 4, the bottom right cell has its form function different than 0
> for dof =0 (angle dof) (i.e *fe_face.shape_value(dof = 0; point_gauss) =/= 0* )
> whereas on the fig. 5, the bottom left cell has *fe_face.shape_value(dof = 0;
> point_gauss) *= 0 for all gauss point.
It is difficult to imagine that something this basic doesn't work given how
widely these functions are used. But since you think you know what is
happening, can you strip down your program to as-simple-as-possible a program
that illustrates the issue?
> So the fig.4 is the RHS for one direction beta=(0.14; 0.14) and fig 5. is the
> RHS for the other direction beta=(0.14;-0.14). The BC
This too does not actually look wrong to me, if you consider how Paraview
plots the rhs vector. You can only trust the values in the vertices of that
bottom row of cells, and the two pictures look perfectly symmetric to me.
Best
W.
--
------------------------------------------------------------------------
Wolfgang Bangerth email:
bang...@colostate.edu
www:
http://www.math.colostate.edu/~bangerth/