I hope you can help me with me question. I have been running a test in CFX4.2 where I used a Poiseuille flow. I meshed the tube with an O-grid patched to an H-grid (i.e. Butterfly grid). But the numerical result deviates from the analytical fully developed profile in the block corners, where the two grids are patched together. This gets even more clear when I look at the velocity gradients for the same region.

And now to my question: Is this due to the fact that the control volumes contain a larger angle the closer one gets to the block corners, or is there some other explanation?

(1). A couple of exercises you can do. (2). The first: change the relative size of the two regions, and check the result to see if it is always happen at the same location. (3). The second: reduce the mesh size (especially the one at the corner) and verify the mesh independent solution. That is, will the solution converge to the analytical solution as the mesh size is reduced. (4). If your current solution is not mesh independent, then you should have some kind of errors. It is nothing unusual. (5). It is important to know the total number of mesh points needed to get the mesh independent solution. (6). There is no way one can evaluate the accuracy of solution based on the finite size mesh. (unless the solution is constant or linear.)

This is a very common meshing practice - if the discretisation is done properly, then you should not see what you saw. Ask the CFX support engineers to see they can fix for you.