SmoothMeshLines() strange behaviour

[isbarak]

openEMS ver 0.0.35 built from source on AlmaLinux 9 kernel 5.14.0-570.24.1.el9_6.x86_64
I am using the Python interface. I have a mesh created by multiple AddLine (). The result before smoothing is:
lx=mesh.GetLines('x',do_sort=True); lx array([ 0. , 29. , 30. , 40. , 41. , 195. , 196. , 392.11111111,
393.11111111, 589.22222222, 590.22222222, 786.33333333, 787.33333333, 983.44444444, 984.44444444, 1180.55555556,
1181.55555556, 1377.66666667, 1378.66666667, 1574.77777778, 1575.77777778, 1771.88888889, 1772.88888889, 1960. ,
1969. , 1970. , 2000. ])
I then perform mesh.SmoothMeshLines('x', resolution, ratio = 1.5) with resolution 312.28381041666665
the new mesh in 'x' direction is unchanged, and obviously does not meat the ratio=1.5 requirement

any advice most welocme

thanks

i

[AJ528]

If those are your mesh lines and your resolution is 312, I wouldn't expect SmoothMeshLines to do anything since there aren't any gaps greater than 312 in your mesh. SmoothMeshLines looks for gaps greater than the mesh resolution and adds mesh lines until the gaps are filled.
If you have a 2 mesh lines close to each other (like 10 apart) and then a huge gap of 1000, SmoothMeshLines doesn't just add 3 lines 312 apart to fill that big gap, it adds lines closer to each other (to better match the 2 that are 10 apart) but slowly grows the gap in lines it puts down until it reaches the mesh resolution. The ratio controls how quickly it goes from placing lines 10 apart to 312 apart.

[isbarak]

For a ratio of 1.5 and given lines at 29.0, 30.0, its should have added lines at 27.5 , 26.75 , 25.625 , 23.9375 ,
21.40625 , 17.609375 , 11.9140625 ...
anyway, I have written a python implementation of smoothmesh() and would be happy to share it

[thliebig]

We should fix any issue with the existing SmoothMeshLines method and not create new ones. I had that with the Matlab/Octave interface... it was/is a mess...

I still think that you have not understood how this is supposed to work? As @AJ528 said, with a requested resolution of 312 and the lines given, nothing should happen...

[isbarak]

Thanks Thorsen for intervening here, and may I say that you have created a very impressive package. Kudus.
Perhaps we should first establish what SmoothMeshLines() should do.
I am suggesting (modifying from the manual): SmoothMeshLines should enables us to refine existing meshes by adding the minimal number of points that will guarantee that the distance between any two adjacent points is less than a predifined distance, and that the ratio of adjacent distances is less than a predefined ratio
once we agree on if and what needs to be improved, I will be happy to cotribute what I have already learned

[thliebig]

I think the definition is mostly correct, the predefined target distance is what resolution stands for...
The ratio is a bit more tricky, it only applies if larger gaps needs to be filled to give a hint on how fast the algorithm can increase from a small to max/given resolution. If a gap is already too small, it should never add lines, even if the ratios are not met...
In this case the user has already "messed" up and needs to find a way to fix it, e.g. asking for a lower resolution in certain areas...

[isbarak]

Thorsen, is the following correct?
1.The limit rate of gap increase should only apply for grid gaps larger than a predefined value
2. This predefined value relates to the simulation timestep, which is set by the smallest gap.
3. in many cases this smallest gap is dictated by some physical or geometrical limit, e.g. transmissin line width, skin depth, capacitor plate distance, etc

if the above is correct, please specify the governing equations
thanks

ilan

[tdunning]

After reading this discussion and taking in the points made here, I went back in and remeshed my latest project with the following method. My results were very good this time which is a contrast with my previous results.

First, I defined three scale numbers. The first, m1, is my shortest wavelength / 20 for free space meshing. The second, m2, is my trace width / 5 for the mesh size near my traces. The third, m3, is either half or a third of my trace mesh size which is what I use to position the grid near the edge of traces or near my ground plane. This should result in m1 > m2 > m3.

Second, I set a pair of mesh lines near the edges of traces using the 1/3, 2/3 rule with one line inside and one outside the trace. If one of these mesh lines was very close to the coordinates of a probe or signal injection port, I omit doing that mesh line explicitly or move the port a bit away. This isn't much of a problem given how tight the edge mesh is at the edges. Previously, I had fits with this.

At this point, I use smooth meshing to fill in at my trace meshing interval, m2. This causes smoothly variable mesh widths as the mesh moves away from the trace edges.

Finally, I add mesh lines outside my simulation area and fill smoothly with my free space interval.

The overall result is both esthetic and seems to avoid the surprises I have had so far. Thanks to the participants in this discussion.