Element imprint on Q-Criterion Isosurfaces

354 views
Skip to first unread message

Alex Tsolovikos

unread,
Mar 2, 2022, 6:48:17 PM3/2/22
to Nek5000
Hi,

I am working on a zero pressure gradient turbulent boundary layer and I'm interested in plotting Q criterion isosurfaces. However, when I compute the Q criterion with paraview, there is a clear discontinuity of the isosurfaces between elements (see attached image).

The simulation appears to be well-resolved (Δx+, Δz+ < 5, Δy+ of first node < 0.1 and it matches the statistics of other similar DNS simulations). 

I know that there is a gradient discontinuity between elements that causes this issue. How can I reduce its effect, besides increasing the grid size? Would increasing the polynomial order (rather than num of elements) help? Is there any smoothing filter in paraview that I can use?

Thanks in advance,
Alex
tbl.png

Fischer, Paul

unread,
Mar 2, 2022, 7:55:30 PM3/2/22
to Alex Tsolovikos, Nek5000

Hi Alex,

I don't have an answer yet but, out of curiosity, what order are you
currently using?

Paul


From: nek...@googlegroups.com <nek...@googlegroups.com> on behalf of Alex Tsolovikos <atsol...@gmail.com>
Sent: Wednesday, March 2, 2022 5:48 PM
To: Nek5000 <nek...@googlegroups.com>
Subject: [nek5000] Element imprint on Q-Criterion Isosurfaces
 
--
You received this message because you are subscribed to the Google Groups "Nek5000" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nek5000+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nek5000/37a91a01-893d-4cf0-a3fa-3f457d1dca59n%40googlegroups.com.

Alex Tsolovikos

unread,
Mar 2, 2022, 9:46:14 PM3/2/22
to Fischer, Paul, Nek5000
Hi Paul,

I am currently using N = 7.

Alex

Philipp Schlatter

unread,
Mar 3, 2022, 2:49:28 AM3/3/22
to nek...@googlegroups.com

Hi,

indeed, this is a problem that I have seen also in the past, in various codes/methods. What I typically did (in spectral codes) was to improve the physical-space representation by (high-order) interpolation. For instance, for the video here (https://www.youtube.com/watch?v=dWe3fnfo9WQ&t=0s) I doubled the resolution in physical space in each direction by spectral interpolation prior to going to VTK.

Another thing that helps a bit is to use the loop subdivision filter in paraview (not sure whether Visit has something similar).

Finally, for SEM-type meshes I would image that interpolating to a uniform mesh prior to VTK may also help to avoid the (linear) interpolation on non-equidistant meshes for the various filters.

What I doubt will help is simple element-based filters/smoothing.

Philipp

Alex Tsolovikos

unread,
Mar 3, 2022, 4:39:04 PM3/3/22
to Nek5000
Hi Philipp,

Thanks for the tips. I interpolated to a uniform grid and then computed Q-criterion and that seems to do the trick! (See attached.)

Thanks,
Alex
q_criterion_before.png
q_criterion_fixed.png

Philipp Schlatter

unread,
Mar 4, 2022, 3:19:15 PM3/4/22
to nek...@googlegroups.com
Great!
Just to be sure: Did you compute Q before or after interpolation on a
uniform grid? If you did it afterwards; how did you compute Q on the
uniform mesh?

Philipp
>> <https://groups.google.com/d/msgid/nek5000/37a91a01-893d-4cf0-a3fa-3f457d1dca59n%40googlegroups.com?utm_medium=email&utm_source=footer>.
>
> --
> You received this message because you are subscribed to the Google
> Groups "Nek5000" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to nek5000+u...@googlegroups.com
> <mailto:nek5000+u...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/nek5000/35fcd2e3-e6e5-4c39-ac83-db75af60277dn%40googlegroups.com
> <https://groups.google.com/d/msgid/nek5000/35fcd2e3-e6e5-4c39-ac83-db75af60277dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

Alex Tsolovikos

unread,
Mar 9, 2022, 11:52:38 AM3/9/22
to Nek5000
Hi Philipp,

I first interpolate the velocity field to a uniform grid and then compute the Q criterion with the interpolated velocity field.

I do both in Paraview as follows:
1) Create a FastUniformGrid (with the desired num of grid points)
2) Transform the FastUniformGrid to the desired size
3) ResampleWithDataset from the original Nek5000 grid to the transformed FastUniformGrid
4) Use the Gradient filter to compute the Q criterion

I also tried exporting the Nek5000 directly to a uniform grid by setting ifreguo = .true., but that did not help get rid of the element imprint.

Is there any other way to interpolate to a uniform grid in Nek5000 besides setting ifreguo = .true. or using history points?

Best,
Alex
Reply all
Reply to author
Forward
0 new messages