Discussion Forum

I opened your file and saw the mesh1 you had in there. When I meshed it, it came up with 622948 elements. That is a big problem to solve.

In the model, I added another mesh and just did a default free tet with default size. It meshed fine and had under 35000 elements. I tried to solve the problem as you had the physics set up but the solution did not converge. I am not all that familiar with laminar flow problems so I plead ignorance there. Try adding in another mesh with a simpler setup. I have an image with the new mesh attached.

Thanks for your reply. I see what you mean, however, the mesh is just boundary mesh, not for the domain. When I meshed it as you said, I saw failure to find a solution and it showed to me that:

Number of vertex elements: 8
Number of edge elements: 225
Number of boundary elements: 5078
Warning: No mesh on domain 1
in geometry 1, mesh case 0.
Warning: Not all domains are meshed.
Warning: No mesh on domain 1
in geometry 1, mesh case 1.
Warning: Not all domains are meshed.
Linear solver
Number of degrees of freedom solved for: 36424.
Nonsymmetric matrix found.

Which means there is no mesh on domain. Clearly, I can not mesh the domain.

Hello Lichun! I have been looking at your model. I too, am interested in the new parametric surface feature. It is really nice and easy to use. Indeed, I think if you set that part up a little differently, it will fix your problem.

You had set it up with 100 knots and a very small convergence criteria near zero (1.0e-18 or something like that). Then COMSOL gave you a warning message that it could not converge lower than another larger number (.001 or similar if I recall).

So, what I did is at 100 knots, I raised the convergence criteria to be slightly larger than the value from the warning message, and I get a much smoother, but less accurate surface. This surface can be meshed easily and solved; I did both.

The next step is to increase the number of knots to be more accurate. I am now rerunning your case with 200 knots.

Keep doing this until you have an acceptable representation of the surface that will also mesh and solve.

You will find that if the number of knots gets too large, it is too large to fit in your computer memory. So, you have to be smart about it.

let us know how it works out !

Dear James. I did what you said of using 100 knots and increased the converge criteria to be large enough to eliminate warning. However it still can not be solved as before and Comsol told me I have no mesh in domain. Could you give me you successful model that has been solved? Thanks very much!

I am not getting the errors you are getting. For me, the problem errors out for convergence. Here is what I get for errors:

Error:

Failed to find a solution.
Maximum number of Newton iterations reached.
Returned solution is not converged.

Warnings (3 of them)

Ill-conditioned preconditioner. Increase factor in error estimate to 68.4076754783063.

Ill-conditioned preconditioner. Increase factor in error estimate to 48.54172082586452.

Ill-conditioned preconditioner. Increase factor in error estimate to 35.42642480261303.

When it comes to the mesh itself, I do not get any domain errors. Perhaps I just don't know where to look but when I right click for mesh statistics, things seem ok.

See image for mesh statistics I see.


I also included the file I have been messing with. Done with ver 4.2.0.187 (the latest).

I find this problem interesting because, while I may not understand it all, just the fact that you are getting different result for the mesh piques my curiosity. You will have to resolve the file because I deleted all of the solutions.

I cannot send you the model file with all the data because it is too large. However, I can send you the model with mesh, solution, and history reset which makes a smaller file size. I am attaching pictures of the mesh and solution to show what you should get. then I started to create the more detailed geomety (increase knots and decrease residual criteria), you can then see the details in the geometry show up. This will require a much more detail mesh. You will need a very large computer to capture all this detail. You elements will need to be very small. You can see how many elements you need for the smoother representation. It works very well, but your data contains a lot of detail that requires detailed elements.

I could not e-mail the 2nd model to you because it is too large.