Freezing error during large mesh generation about py-tdgl HOT 3 CLOSED

Wow thanks so much!

from py-tdgl.

Hi Owen,

I've opened a Pull Request (#20) which optimizes the process of creating the necessary data structures after mesh generation. I've also added a progress bar to the parts of the process that take the longest.

With the new code, I get the results below. I am using max_edge_length = xi because max_edge_length = xi/2 results in a mesh with over a million sites. Even with max_edge_length = xi, the mean edge length is roughly xi / 2, which is probably good enough.

The other thing to watch out for when making really large meshes is the number of points in the underlying tdgl.Polygon. You may have to play around with the number of points in the Polygon using tdgl.Polygon.resample() and tdgl.Polygon.buffer() to get rid of the Malformed Voronoi cell surrounding boundary site warnings.

The other error I am seeing in your logs is RuntimeWarning: divide by zero encountered in double_scalars ... current_density = (-1 / term.length) .... This means that one of the tdgl.Polygon instances you are using to define a transport terminal doesn't actually contain any boundary mesh sites.

I will merge this PR and push the changes to PyPI later today or tomorrow. In the meantime, if you'd like you can check out the branch that has the changes: https://github.com/loganbvh/py-tdgl/tree/big-mesh

Let me know if you have any other questions!

import logging
logging.basicConfig(level=logging.INFO)

import tdgl
from tdgl.geometry import box

xi = 0.005
d = 0.01
london_lambda = 0.2

layer = tdgl.Layer(coherence_length=xi, london_lambda=london_lambda, thickness=d)
film = tdgl.Polygon("film", points=box(2, 1)).resample(5001).buffer(0)
device = tdgl.Device("bar", layer=layer, film=film, length_units="um")

device.make_mesh(max_edge_length=xi)

Output:

INFO:tdgl.device.device:Generating mesh...
INFO:tdgl.device.meshing:Iteration 1: 23153 points, 41291 triangles, max_edge_length: 2.2e-01 (target: 5.0e-03).
...
INFO:tdgl.device.meshing:Iteration 22: 298368 points, 591716 triangles, max_edge_length: 4.9e-03 (target: 5.0e-03).
Finding Voronoi indices: 100%|████████████████████████████████████████████| 1775148/1775148 [00:14<00:00, 122284.12it/s]
Constructing Voronoi polygons: 100%|█████████████████████████████████████████| 298368/298368 [00:17<00:00, 16874.98it/s]
INFO:tdgl.device.device:Finished generating mesh with 298368 points and 591716 triangles.
INFO:tdgl.device.device:Total mesh generation time: 79.232 seconds

device.mesh_stats_dict():

{'num_sites': 298368,
 'num_elements': 591716,
 'min_edge_length': 0.0009765625,
 'max_edge_length': 0.00494772146487779,
 'mean_edge_length': 0.0028552928379728393,
 'min_area': 2.900378605089071e-07,
 'max_area': 1.2706923858807462e-05,
 'mean_area': 6.7031299058204325e-06,
 'coherence_length': 0.005,
 'length_units': 'um'}

from py-tdgl.

With some further optimization, the total mesh generation time for this model is about 40 seconds. About half of this time is spent converting MeshPy C++ arrays to numpy arrays - not much I can do about that. The other half of the time is mostly taken up by finding the areas of the Voronoi cells and the lengths of the Voronoi edges. This can probably be sped up further, but I don't see any more obvious low-hanging fruit right now.

INFO:tdgl.device.device:Generating mesh...
INFO:tdgl.device.meshing:Iteration 1: 23153 points, 41291 triangles, max_edge_length: 2.2e-01 (target: 5.0e-03).
...
INFO:tdgl.device.meshing:Iteration 22: 298368 points, 591716 triangles, max_edge_length: 4.9e-03 (target: 5.0e-03).
INFO:tdgl.device.device:Creating Mesh object from triangulation.
Constructing Voronoi polygons: 100%|█████████████████████████████████████████| 298368/298368 [00:11<00:00, 25605.58it/s]
INFO:tdgl.device.device:Finished generating mesh with 298368 points and 591716 triangles.
INFO:tdgl.device.device:Total mesh generation time: 41.446 seconds

Merging #20 into main

from py-tdgl.