Hi all,

can any body tell me the criteria to determine the mesh size and time step size in hand before solving the problem

is there any estimate where in we can say that we can go till this paritcular size for mesh and time step.

how are these dependent upon the initial and boundary conditions.

we know that high frequiencies are always solved on fine mesh sizes and so we determine once we know the highest frequency in the problem.

thanks in advance

I always did a grid check. I got a solution that I was happy with, and then I would increase the mesh count and see if the answer was the same or at least within some %, and then I would also decrease the mesh size and check the answer again. If the answered changed substantially when I increased the mesh size I knew something was wrong and would increase the mesh count again and see how that answer compared. You can visually show this in a graph of mesh count v's answer..you will flat line after you find the least amount of cells

Hi

Thanks for the reply

but thats a crude way of setting the size.

in first hand can we try to predict by knowing the frequency range ofthe problem under cosideration.

If you're running a non-steady simulation (i.e. real transient or pseudo-transient), than your mesh size and time step are linked together via Courant number. Though there are strict restrictions only explicit schemes, which are not so widely used, even for implicit scheme putting a Courant number above 500...1000 will make no good.

Another limitation comes from y+ if you do a turbulent simulation (and if it's not DNS). This can give you the size of near-wall cells. And you can actually derive the size of all other cells from mesh quality considerations like aspect ratio limit and also the fact that most differencing schemes doesn't perform well if two adjacent cells differ in size more than 1.2...1.5 times.