Hello,

I'm simulating a flow past cylinder using an O grid, in FVM. I normally intialize solution everywhere to u=1. and do the simulation. As time step increases, the velocity near the boundary will decrease since u=v=0 on its surface. It works ok when 1st grid size is at 5.e-3.

When I refine my grid further to 1.e-3, 2 strange things happen to the grids near the boundary. The velocity decreases as it gets nearer the boundary but for the grid cells just next to the boundary, velocity suddenly gets larger and points in the opposite direction. Hence the resulting drag is negative.

Interestingly the solution did not explode and other parts of the flow looks fine (separation etc). At 5.e-3, there's no such problem. I also tried to initialize the solution such that near the boundary, the velocity is tangential to it and it gets smaller as it get nearer but the problem is still there.

Moreover, I've used the code on a C grid simulating flow past an airfoil and there's no such problem even as grid refines. Hope someone can suggest some ways to solve this problem.

Thank you.

What precision are you using? Although 1e-3 doesn't seem too small, it might be worthwhile to try double precision to see the effect.

I'm using double precision already. Btw, I wonder if it is because for initial velocity field is not solenoid. I'm not very sure how to do it. Can someone enlighten me? Thanks.

Why not just use a uniform flow field as the initial condition? This is divergence free.