AMR error after Stokes solve.

[bknight1]

Unable to adapt mesh with variable (e.g. strain rate) after stokes solve. Most likely due to variable not being the velocity or pressure field. I'm currently on the latest development branch with PETSc v2.24.4 that is built with pragmatic, mmg & parmmg. Here is the error message:

[0]PETSC ERROR: --------------------- Error Message --------------------------------------------------------------
[0]PETSC ERROR: Argument out of range
[0]PETSC ERROR: Invalid field number 2; not in [0, 2)
[0]PETSC ERROR: WARNING! There are unused option(s) set! Could be the program crashed before usage or a spelling mistake, etc!

[lmoresi]

Can you add more information so we can reproduce this - we're continually debugging this part of the code which is difficult to get right and changes quite a bit with petsc updates.

[bknight1]

Here's a basic example using the Stokes sinker, adapting based on the SR field:

# %%
import sympy
import underworld3 as uw
from underworld3 import meshing, discretisation, systems, constitutive_models, adaptivity

# %%
mesh = meshing.UnstructuredSimplexBox(
    cellSize=0.02,
    minCoords=(0.0, 0.0),
    maxCoords=(1.0, 1.0),
    qdegree=2,
)

# %%
v = discretisation.MeshVariable("U", mesh, mesh.dim, vtype=uw.VarType.VECTOR, degree=2)
p = discretisation.MeshVariable("P", mesh, 1, vtype=uw.VarType.SCALAR, degree=1)
sr = discretisation.MeshVariable("SR", mesh, 1, vtype=uw.VarType.SCALAR, degree=1)

uw.pprint(0, "Setup basic Stokes sinker model")
# %%
x, y = mesh.X
x0, y0, r = 0.5, 0.7, 0.15
rho_light, rho_heavy = 0, 1
g = 1

# %%
rho = sympy.Piecewise(
    (rho_heavy, (x - x0) ** 2 + (y - y0) ** 2 < r**2),
    (rho_light, True),
)

# %%
stokes = systems.Stokes(mesh, velocityField=v, pressureField=p)
stokes.constitutive_model = constitutive_models.ViscousFlowModel
stokes.bodyforce = sympy.Matrix([0, -rho * g])

# %%
for wall in ("Left", "Right", "Bottom", "Top"):
    stokes.add_dirichlet_bc((0.0, 0.0), wall)

# %%
uw.pprint(0, "Solve basic Stokes sinker model")

# %%
stokes.petsc_options["ksp_monitor"] = None
stokes.solve()

# %%
uw.pprint(0, "Save output")

# %%
sr.data[:, 0] = uw.function.evaluate(stokes.Unknowns.Einv2, sr.coords).reshape(-1)

# %%
mesh.petsc_save_checkpoint(0, [sr, v, p] )

# %%
uw.pprint(0, "Begin mesh adaptation")


# %%
metric = adaptivity.metric_from_field(sr, h_min=0.01, h_max=0.1)
mesh.adapt(metric)


# %%
uw.pprint(0, "Finish mesh adaptation")

[lmoresi]

Fixed in PR #67 (merged to development). The adapt() method now destroys all old variable lvecs upfront before reinitializing, preventing stale field_id errors in _setup_ds(). The metric variable is also preserved through adaptation.

[bknight1]

Confirmed working. Thanks, Louis