Dear all, I'm interesting in unstructured grid. Please tell me how to start. thanks for your help Wee. Thailand

Then tell us more... What do you want to do? Implement your own meshing? Find out general info about unstructured meshing in CFD? Is it for a particular problem?

Dear Herve Just general info....I think. Cause now I don't have any idea about it. But I see it in many papers. So I wanna know what it is. thanks for help Wee.

(1). The mesh like a graph paper is structured mesh, because the grid point is addressed like (i,j) in an organized fashion. (2). The unstructured mesh is one which does not have such method of address. In other words, grid(1), grid(2), grid(3),etc... can be anywhere in the computational domain.Like a movie theater, you can sit anywhere.

What about the convergence if we sit everywhere? Can we have method to optimize so that get the best and fastest convergence?

(1). Good question. (2). So, the unstructured mesh is not designed to get faster convergence. It is designed to get a quick mesh. (3). It is a good idea to avoid unstructured mesh when you are interested in faster convergence. (4). Arbitrary distribution of mesh (indexing) can only create more resistance like insulator material. Structured mesh will provide less resistance and and better convergence, in general.

John,

Unstructured meshing also offers a lot more ease to mesh complex geometries. I was looking at natural river/estuarine geometries and structured meshing was very difficult indeed. Unstructured meshing offered an easier numerical epresentation of the domain. Now it is true that one has (I feel) more control over the mesh and numerical convergence when using a structured grid; yet if the very nature of the grid is detrimental to convergence (esp. for turbulent terms, maybe it is worth switching to a less performant grid on which convergence properties although slower will be better. Regarding quick mesh I agree with John, but it does not mean "good" mesh. Most commercial code offer both approaches (CFX, Fluent).

(1). Good point. (2). The issue here is, we sometimes are very naive to think that a problem with a complex geometry can be handled just like the one with simple geometry. (3). As a matter of fact, the solution to a problem with simple geometry most of the time is very complex. (4). If we look at the problem with complex geometry (which will give complex solution) and then divide it into smaller but managable problems (and computational domains), then each sub-problem should be easier to handle in terms of mesh generation and solution. (5). These sub-problems can be solved decoupled, just like old fashioned inviscid/viscous boundary layer problem. (6). Or it can be solved coupled through "multi-blocked structured mesh", if it is hard to define boundary conditions separately for each sub-problem. (7). It is a common sense that a family doctor can treat only ordinary sickness for short time. But sometimes, people just don't want to go into the trouble to see a specialist. (8). Actually, for the finite number of mesh points, there is an optimum arrangement of mesh to get the best solution. But beyond that, the mesh independent solution does not care how the mesh is created at all. (if you know how to get the solution to converge!)