decimation not working as expected.

Question

decimation not working as expected.

mdforti opened this issue 3 years ago · 10 comments

I am sorry if this is not the place but I can not create a new account at the support forum:

New registrations are not allowed from your IP address (maximum limit reached). Contact a staff member.

I need to ask something about the surface smoothing and increasing number of faces in the surface. My objective is to smooth the surface at the plot without open3d.
I cant get the surface smooth and decimator_method to work. whatever values I use for thee options the results are allways the same.
at some point I got a warning for 'decimation need open3d' but after installing this requirement I get planar surfaces as shown in the figure 'with open3d', instead of expected result. I would appreciate a clue about to where to start debugging.

whithout open3d:

with open3d:

Answer 1 · 2021-08-12T10:58:22.000Z

Hi @mdforti, can I just check what version of IFermi you're using? I fixed some bugs with decimation in versions 0.2.4 and 0.2.5 released earlier this year.

Answer 2 · 2021-08-12T11:33:27.000Z

pip says it installed 0.3.0. there is a big chance I am making the mistakes also. also note that to be able to install open3d I had to use python 3.8 which can also be causing issues.

Answer 3 · 2021-08-12T12:10:33.000Z

Ok. To get smoother surfaces you should try increasing the interpolation factor for Fourier interpolation. Decimation will do the opposite - it will make the surfaces less smooth (fewer faces). The smoothing option isn't well tested. The preferred option is to just increase the interpolation factor.

Answer 4 · 2021-08-12T12:12:26.000Z

For example, the README example code is:

interpolator = FourierInterpolator(bs)
dense_bs = interpolator.interpolate_bands()

This could be changed to:

interpolator = FourierInterpolator(bs)
dense_bs = interpolator.interpolate_bands(interpolation_factor=50)

You can try increasing the interpolation factor even more than that.

Please let me know if this works for you.

Answer 5 · 2021-08-12T12:19:10.000Z

Aaah ok I was getting decimation completely wrong then. I was interpreting the help, which says 'if decimation factor is a float then it si the factor to reduce the number of faces', then tested decimation factors < 1. now I see it.
thank you very much for the clarification !

Answer 6 · 2021-08-12T12:20:26.000Z

Great! Do you have smooth surfaces now?

Answer 7 · 2021-08-12T12:28:38.000Z

yes, its working now !
however to tell you the truth, interpolation factor beyound 10 really puts strain on my laptop which is an i7. what might I be doing wrong ?
just tested: interpolation factor of 15 crashes the execution.

Answer 8 · 2021-08-12T12:39:59.000Z

Hmm if your band structure has a lot of bands then it may be doing a lot of unnecessary interpolation (only bands near the valence/conduction/Fermi level need to be included). In that case you can use the trim_bandstructure function which will remove excess bands: https://fermisurfaces.github.io/IFermi/ifermi.html#ifermi.interpolate.trim_bandstructure. E.g.,

from ifermi.interpolate import trim_bandstructure

bs_trim = trim_bandstructure(1.5, bs)
interpolator = FourierInterpolator(bs_trim)
dense_bs = interpolator.interpolate_bands(interpolation_factor=15)

That should make the interpolation easier.

Is interpolation_factor=10 still very blocky?

Answer 9 · 2021-08-12T12:44:27.000Z

you are partly correct, I have many bands above the cbm and considerable amount below vbm so I guess I can remove them.

Answer 10 · 2021-08-12T13:48:26.000Z

I'm closing this issue but feel free to re-open if you have more questions.