CFD Online Discussion Forums - Curvilinear Coordinates

Recently I converted my orginal FVM code, a very neatly and well verified code that solves a flow problem on a cartesian cordinate system, to a generalized non-orthogonal coordiante system. I have put some changes and I am sure all the curvilinear features added are correctly made. The main idea of my effort is to solve the flow problem on a transformed uniform grid, in which the physical problem was set on a stretched grid.

For simplicity I solved the scalars on the physical coordinate. With the curvilinear feature added, I witnessed the code has become powerful, with less number of grids and few pressure correction iteration, for a high Re flows, converges relatively at ease. However, the solution is dependent on the grid stretching parameter, and is only the same to the benchmark solution, with the old code, when only run on a uniform grid.

I have noticed one of this forum active member, Harish, have at once posted a similar problem to mine, some years back, (right now I can't exactly locate the link).

Can anyone or Harish be of help on this matter. I spend almost a week to resolve this, with no avail.

My assumption is if the scalar quantities are solved on the physical plane they don't need any inverse transformation. Inverse transform would be needed if I solved the flux vectors , or covariant or contra variant vectors.

Finally, if some one wants to discuss how I implemented the problem, the governing equations, discretization, etc please email me on tawnhb@yahoo.com. But I used collocated grid, 3d, solve buoyancy driven flow, fractional step, multigrid ( and/or bicgsatb solver for pc, etc). Indeed a good code.

I appreciate your enlightenments on my difficulty.

Ps I have only average flare on tensors, but I generally understand read was able to grip and implement the ideas of some paper by {Yan Zang, RL Street.. and another one by Melaaen, and they were very informative.

With best regards,

TAW