I am trying to solve incompressible NS equations iteratively using NGSolve. I am ignoring the incompressibility condition (using a splitting method). Hence, what I am trying to solve has a non-linear convection term on the l.h.s and a known pressure value on the r.h.s and I am solving for velocity, iteratively.

I have taken a look at the documentation the section regarding the non-linear equations and solving them using Newton's method. I put the pressure term on the l.h.s of the equation and got Ax=0 type equation and followed Newton's method as explained. But I am not converging and I do not know why..

Anyone has experience with this?

Thanks

Here I write the equations in the way that I want to solve them. (\nu=viscosity)

u^{t+1}-u^{t}/deltat - \nu*1/2*(\laplacian u^{t+1}) + 1/2*(\grad u{t+1} * u^{t}) = - \grad p^{t}+ \nu*1/2*(\laplacian u^{t}) - 1/2*(\grad u{t} * u^{t})]]>

I want to generate triangular mesh for concentric cylinder with a thin membrane between inner and outer region:

My input file :

algebraic3d

solid cube1 = orthobrick (0, 0, 0; 1400, 1400, 800);

solid cyl1 = cylinder (700, 700, 0; 700, 700, 800; 200.0);

solid dom1=cyl1 and cube1;

solid cube2 = orthobrick (0, 0, 0; 1400, 1400, 800);

solid cyl2 = cylinder (700, 700, 0; 700, 700, 800; 210);

solid gab2= cyl2 and cube2;

solid dom2= gab2 and not dom1;

solid cube3 = orthobrick (0, 0, 0; 1400, 1400, 800);

solid cyl3 = cylinder (700, 700, 0; 700, 700, 800; 700);

solid gab3= cyl3 and cube3;

solid dom3= gab3 and not dom1 and not dom2;

tlo dom1 -col=[1,0,0];

tlo dom2 -col=[0,1,0];

tlo dom3 -col=[0,1,0]- transparent;

The coarse mesh looks like this:

drive.google.com/open?id=1FX0AJ71svwhZC9cd4WlG-6dURP6enVe5

and to simulate the "physics" that iam interested in, i need to refine the mesh near the membrane, so it looks like this:

drive.google.com/open?id=1LISFB0zoEPibIXDX4y57ZOZ7urS03UeG

The number of mesh is too large and i know the change of physics quantity iam interested is signifiicant only in radial direction. Is there a way to refine mesh only in particular direction, in my case in radial direction? This way i can save lots of computational time and resource.

(

I could make more multiple regions and mesh them later to look like this :

drive.google.com/open?id=10nY9n2CtW4su6HPLca9BJ32DmEZ659Qr

But when i tried to increase the length of cylinder, the mesh generation took forever)

Thanks,

Alex

[attachment=undefined]coarse.png[/attachment] ]]>

I am modifying a 2D time-domain PML code (see attached by Joachim) to incorporate the existence of evanescent waves.

In some other problems, there exist evanescent waves. In this case I want to set the pml scaling parameter has a negative imaginary part to accelerate the decay of evanescent waves. Does the pml support this feature?

Best,

Shixu]]>

I am trying to build the latest Netgen/NGSolve from source on MacOS and following the instruction of the website I get an error when its downloading "project_tkdnd"(I have attached the error log and also a screenshot of the terminal).

I know that using the DMG to install works fine on this system and also Ive managed to compile from source on a different linux machine. So, I am not able to see an immediate problem with the installation.

Please let me know where I am going wrong with the installation.

That being said, the reason I current build from source instead of using the DMG is because I need a few more grid function templates instead of the default ones.

Best regards

Ajay]]>

is there a good way to compute the (element-wise) L

Best wishes,

Henry]]>

i'm trying to implement a weighted residual-based error estimator for a 2D Poisson equation. It consists of an inner and an edge part. Similar to the one Philip Lederer implemented in the showcase

ngsolve.org/showcases/ngsolve/9-implemen...-a-lifting-technique

My problem is, that the weight is given by (h_T/p_T)^2 for the elements and h_e/p_e for the individual edges.

The first part is no problem since Joachim added the GetOrder()-function to the python interface.

I tackle the edge part by

+ Defining a bilinear form with the .other()-function

+ Using Apply

+ Integrate element_wise

Is this the correct way of doing this? (Without using the lifting approach)

And furthermore how can i incorporate the p_e for an individual edge?

Best regards,

Tim]]>

I have a question about the domains of integration. The example [Spaces and forms on subdomains] shows that the integration domains can be boundary regions:

Can be boundary regions of the grid nodes or points of geometry?]]>

Is it possible to read in mesh previously generated with netgen and use them with the Python script solver ngs-py?

In the .vol file I know what the ids of the different materials will be and the ids of the surfaces are, but these are tagged just with 1,2,3 etc rather than text labels so I’m also not sure how to prescribe the materials and the BCs as the example shown uses meshes generated within the Python script solver which have textual labels instead.

Also, I’d be grateful if you could update the build of netgen/ngsolve for a MAC as I think this still has the bug of only generating meshes for a sphere of unit radius c.f.

ngsolve.org/forum/ngspy-forum/666-bug-meshing-ball-in-3d#1248

Thanks

Ben]]>