[AMRVAC] bug: downsampling AMR is broken

[neutrinoceros]

Bug report

based off https://yt-project.org/docs/dev/cookbook/calculating_information.html#downsampling-an-amr-dataset

I noticed that this feature doesn't work as intended at least for AMRVAC data.

working example with a fake amr dataset

import yt

ds = yt.testing.fake_amr_ds(fields=["density"])
ds.print_stats()
p1 = yt.SlicePlot(ds, "z", "density")
p1.save("/tmp/fake_ds_raw")

ds.max_level = 1
p2 = yt.SlicePlot(ds, "z", "density")
p2.save("/tmp/fake_ds_dowsampled")

output

level   # grids        # cells       # cells^3
----------------------------------------------
  0      1           32768              32
  1      8           34304              33
  2      8          181888              57
  3      8          646968              87
  4     15          947856              99
----------------------------------------------
        40         1843784

t = 0.00000000e+00 = 0.00000000e+00 s = 0.00000000e+00 years

Smallest Cell:
    Width: 6.330e-28 Mpc
    Width: 6.330e-22 pc
    Width: 1.306e-16 AU
    Width: 1.953e-03 cm

broken example with AMRVAC

import yt

ds = yt.load("amrvac/bw_3d0000.dat")
ds.print_stats()
p1 = yt.SlicePlot(ds, "z", "density")
p1.save("/tmp/amrvac_ds_raw")

ds.max_level = 1
p2 = yt.SlicePlot(ds, "z", "density")
p2.save("/tmp/armvac_ds_downsampled")

output

level   # grids        # cells       # cells^3
----------------------------------------------
  1     56           28672              31
  2     64           32768              32
----------------------------------------------
       120           61440

t = 2.00000000e-02 = 1.71746294e-06 s = 5.44231165e-14 years

Smallest Cell:
    Width: 1.013e-26 Mpc
    Width: 1.013e-20 pc
    Width: 2.089e-15 AU
    Width: 3.125e-02 cm

[matthewturk]

Yup that looks broken to me, too. My suspicion is that it has something to do with missing non-leaf node data or the level 1 offset.

On Tue, Mar 10, 2020 at 8:32 AM Clément Robert notifications@github.com wrote:

Bug report

based off

https://yt-project.org/docs/dev/cookbook/calculating_information.html#downsampling-an-amr-dataset

I noticed that this feature doesn't work as intended at least for AMRVAC data.

working example with a fake amr dataset

import yt

ds = yt.testing.fake_amr_ds(fields=["density"]) ds.print_stats() ad = ds.all_data() p1 = yt.SlicePlot(ds, "z", "density") p1.save("/tmp/fake_ds_raw")

ds.max_level = 1 p2 = yt.SlicePlot(ds, "z", "density") p2.save("/tmp/fake_ds_dowsampled")

output

level # grids # cells # cells^3

0 1 32768 32 1 8 34304 33 2 8 181888 57 3 8 646968 87 4 15 947856 99

      40         1843784

t = 0.00000000e+00 = 0.00000000e+00 s = 0.00000000e+00 years

Smallest Cell: Width: 6.330e-28 Mpc Width: 6.330e-22 pc Width: 1.306e-16 AU Width: 1.953e-03 cm

[image: fake_ds_raw_Slice_z_density] https://user-images.githubusercontent.com/14075922/76317063-cf176c00-62db-11ea-9d1e-8a7c25ef0af1.png [image: fake_ds_dowsampled_Slice_z_density] https://user-images.githubusercontent.com/14075922/76317066-d0489900-62db-11ea-9640-fe94981bbf66.png

broken example with AMRVAC

import yt

ds = yt.load("amrvac/bw_3d0000.dat") ds.print_stats() ad = ds.all_data() p1 = yt.SlicePlot(ds, "z", "density") p1.save("/tmp/amrvac_ds_raw")

ds.max_level = 1 p2 = yt.SlicePlot(ds, "z", "density") p2.save("/tmp/armvac_ds_downsampled")

output

level # grids # cells # cells^3

1 56 28672 31 2 64 32768 32

     120           61440

t = 2.00000000e-02 = 1.71746294e-06 s = 5.44231165e-14 years

Smallest Cell: Width: 1.013e-26 Mpc Width: 1.013e-20 pc Width: 2.089e-15 AU Width: 3.125e-02 cm

[image: amrvac_ds_raw_Slice_z_density] https://user-images.githubusercontent.com/14075922/76317086-da6a9780-62db-11ea-8021-ac3d6952bcde.png [image: armvac_ds_downsampled_Slice_z_density] https://user-images.githubusercontent.com/14075922/76317088-db9bc480-62db-11ea-9198-1f95df7ae86b.png

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/yt-project/yt/issues/2479?email_source=notifications&email_token=AAAVXO4YIOJ6MSNIQK4GUOTRGY6PFA5CNFSM4LE672HKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IT4ORTQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAAVXO2NGKOZV2FNE46VML3RGY6PFANCNFSM4LE672HA .

[neutrinoceros]

Sounds likely, as there's no non-leaf node at all in AMRVAC, and I don't think we currently mock anything of the sort.

[neutrinoceros]

I searched for differences between the fakedata example and the amrvac one. All I can say for now in the first case, the method get_global_startindex gets called while AMRVACGrid.get_global_start_index doesn't. I was originally thinking this method is broken but maybe it's more about why it's not getting called. I really don't know how to tackle this, it's even hard to know what portions of yt's core are calling each other.

[n-claes]

Well, AMRVAC only writes data from blocks at the bottom of their respective branch in the AMR tree to the datfile, so all nodes above are empty (for obvious data-storage reasons, the code rebuilds this if needed during a restart). So since a block does not have parents/children containing data (we also never set them in the frontend), I assume that's the reason why downsampling fails? Does yt use data from the parent nodes to downsample?