Diagonal slice of contour plot is jagged

[EdSantilli]

The setup

I have 3D, zone-centered, density data scaled from 0 at the top of the domain to 1 at the bottom of the domain. This data represents a stable stratification with small deviations, so there are no overturns or turbulent regions -- the data is smooth. I want to take a cross-section, 45 degrees in the xy-plane through the diagonal and then look at the contour corresponding to a density of 0.5.

The problem

When I do this using a Contour plot and a Slice operator, the resulting contour is jagged even though the data is smooth. Interestingly, this only happens when my slice is diagonal and not parallel to the coordinate lines.

Here is an image of the problem.

And here is an image of my pipeline and how I sliced the domain.

I've tried slicing a pseudocolor plot with an isosurface operator, as suggested in #10173. The results are identical. I also tried resampling as suggested in several post. Resampling flattens the lines far from the density perturbation, but completely ruins the perturbation itself. which should be a smooth wave of depression.

My data uses AMR, which is why there are multiple contours near the density perturbation. We can ignore that for now. The main problem is that far from the density feature, the contour should be dead flat. Does anyone have an idea of why this is happening?

Thank you.

[markcmiller86]

I think you want to recenter the data from zones to nodes. Use just a pseudocolor plot at first and no other operators but in the pseudocolor plot attributes, you will see a Centering options with original, nodal and zonal as options. Check nodal and hit apply. The doe the slice and contour operators and see if that doesn't improve your resolve your issue.

I believe the issue is that zone-centered data is by its very nature "jagged". It is a piecewise constant function over each zone. Within a zone is it very smooth. Its actually perfectly flat and unvarying. However, as you cross between zones, the underlying "field" (or "function") being numerically represented has a step (or jaggy) in it. This is unavoidable due to the basis functions zone-centered data implies. I imagine the problem is even more pronounced on oblique slices too.

By re-centering to the nodes, you are basically changing it to a node-centered field which then has no "steps" between zones and so looks smoother.

[EdSantilli]

Thanks, Mark.

There is no contour operator that I can apply to a pseudocolor plot. So, to give your suggestion a try, I followed your recommendation with an isosurface operator applied to the sliced pseudocolor plot. Unfortunately, I get the exact same jagged results.

For the record, when I make the pseudocolor plot nodal and apply a resampling operator before the isosurface, here is what I get.

Ed

[markcmiller86]

Hmmm...ok, forget the Centering option in the Pseudocolor plot attributes. Just add Pseudocolor plot and then add Iso-surface operator (found under Operators->Slicing->Isosurface. I think the Iso-surface operator is smart enough to recent zone-centered data to nodes automatically.

[EdSantilli]

Same results. Regardless of the centering used in the pseudocolor plot, I get jagged results with any of these pipelines:

1. pseudocolor -> isosurface -> slice,
2. pseudocolor -> slice -> isosurface, and
3. contour -> slice

Look at the pseudocolor plot in my last post. In that image, used a nodal centering so that VisIt would interpolate within each cell. Away from the depression, the density is a function of z alone, yet the nodal interpolator is producing a function of x'. You can see that the pseudocolor plot itself is jagged. In other words, both the contour and nodal pseudocolor plots produce incorrect results when sliced diagonally. The sliced zonal pseudocolor plot does not do this.

[markcmiller86]

Would it be possible to share the actual data file here? If too big, you can try one of the methods described in our largedata repo.

[EdSantilli]

Certainly. To ease things, I generated a small, 3D file that illustrates the problem. It has 32x32x32 cells, and the data component titled "density" contains a simple function of z. You can pull up pseudocolor --> isosurface and see that all isosurfaces are flat and normal to z.

To see my issue, create a pseudocolor plot of "density." Then, put that through a slice operator with an normal vector <1 -1 0> and an origin at a point <0 0 0>. If you do this, and force nodal data, the data will appear as if it has a strange horizontal dependence. It's even more compelling if you overlay a contour plot with the same slicing.

Here is a picture of what I see from this very data set.
density_diagSlice_JaggedContours.zip

Here is the data set.
density.zip

[markcmiller86]

Ok, thanks for data and pictures of results. I believe my assessement is correct. But, the approaches I have suggested to correct I guess were not appropriate for this dataset.

Time to take out the sledge hammer 🔨.

Use VisIt's expression system to create a new, node-centered version of density using the expression recenter(<density>,"nodal") called density_nodal. Use that variable instead of density. That will force a recentering to the nodes and give you smooth contours as you sand see below

[EdSantilli]

Excellent! Thank worked!
Thank you for your help, Mark.