[BUG] -AttributeError: 'NaturalBC' object has no attribute 'fn_F' when using add_natural_bc

[Belay-129]

Hi uw3 support team,
I encountered an error when applying a natural boundary condition to the Poisson solver in Underworld3, which prevents the solve from running successfully.

poisson.add_dirichlet_bc([17.0], "Top")
q_bottom = 0.065 # W/m2
poisson.add_natural_bc([q_bottom], "Bottom")

poisson.solve()

When applying a natural boundary condition to a Poisson solver, the following error occurs:

AttributeError Traceback (most recent call last)
Cell In[18], line 1
----> 1 poisson.solve()

File src/underworld3/cython/petsc_generic_snes_solvers.pyx:872, in underworld3.cython.generic_solvers.SNES_Scalar.solve()

File src/underworld3/cython/petsc_generic_snes_solvers.pyx:194, in underworld3.cython.generic_solvers.SolverBaseClass._build()

File src/underworld3/cython/petsc_generic_snes_solvers.pyx:744, in underworld3.cython.generic_solvers.SNES_Scalar._setup_pointwise_functions()

AttributeError: 'NaturalBC' object has no attribute 'fn_F'

Thanks for your time and help

[gthyagi]

Hi @Belay-129,

Please attach the example script you are running so we can debug. It also looks like you’re using an older version of Underworld, please update to the latest release, as some of these issues are already fixed.

Cheers,
Thyagi

[Belay-129]

Hi Thyagi, Thank you for your response. Could you please confirm the current version of Underworld? I checked on GitHub and found the version listed as: __version__ = "0.99.0b" This matches the version I already have installed. Cheers, Belay

…

________________________________ From: Thyagarajulu Gollapalli ***@***.***> Sent: Tuesday, September 23, 2025 11:54 AM To: underworldcode/underworld3 ***@***.***> Cc: Belay Mino ***@***.***>; Mention ***@***.***> Subject: Re: [underworldcode/underworld3] [BUG] -AttributeError: 'NaturalBC' object has no attribute 'fn_F' when using add_natural_bc (Issue #25) [https://avatars.githubusercontent.com/u/44185316?s=20&v=4]gthyagi left a comment (underworldcode/underworld3#25)<#25 (comment)> Hi @Belay-129<https://github.com/Belay-129>, Please attach the example script you are running so we can debug. It also looks like you’re using an older version of Underworld, please update to the latest release, as some of these issues are already fixed. Cheers, Thyagi — Reply to this email directly, view it on GitHub<#25 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BR5VEZYTWDBVDZVA7VR4PBT3UCR6XAVCNFSM6AAAAACHEFJ7TKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMRSGA4TQMZRGI>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[Belay-129]

'''
mesh = uw.meshing.UnstructuredSimplexBox(
minCoords=(423755, -10000),
maxCoords=(496255, 0),
cellSize=1000,
regular=False,
)
'''

'''
k_scalar = uw.discretisation.MeshVariable("k", mesh, 1)
'''

'''
T_soln = uw.discretisation.MeshVariable("T", mesh, 1)
'''

'''
Set up the Poisson system
poisson = uw.systems.Poisson(mesh, u_Field=T_soln)

poisson.constitutive_model = uw.constitutive_models.DiffusionModel

poisson.constitutive_model.Parameters.diffusivity = k_scalar.sym[0]

poisson.f = 0
'''

BCs

'''
poisson.add_dirichlet_bc([17.0], "Top")
q_bottom = 0.065 #W/m2
poisson.add_natural_bc([q_bottom], "Bottom")
'''

solve
poisson.solve()

[gthyagi]

Hi Thyagi,
Thank you for your response. Could you please confirm the current version of Underworld? I checked on GitHub and found the version listed as:
version = "0.99.0b"

This matches the version I already have installed.
Cheers,
Belay
…

The version number may look the same, but we’ve recently pushed many important changes.

[jcgraciosa]

Hello @gthyagi and @Belay-129, I found the cause of the issue and I'm investigating a fix.
I'll give you some updates soon.

Cheers,
JC

[Belay-129]

Thank you Juan for updates. Looking forward to hearing from you. Thanks

…

________________________________ From: jcgraciosa ***@***.***> Sent: Tuesday, September 23, 2025 2:12 PM To: underworldcode/underworld3 ***@***.***> Cc: Belay Mino ***@***.***>; Mention ***@***.***> Subject: Re: [underworldcode/underworld3] [BUG] -AttributeError: 'NaturalBC' object has no attribute 'fn_F' when using add_natural_bc (Issue #25) [https://avatars.githubusercontent.com/u/12045001?s=20&v=4]jcgraciosa left a comment (underworldcode/underworld3#25)<#25 (comment)> Hello @gthyagi<https://github.com/gthyagi> and @Belay-129<https://github.com/Belay-129>, I found the cause of the issue and I'm investigating a fix. I'll give you some updates soon. Cheers, JC — Reply to this email directly, view it on GitHub<#25 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BR5VEZ5B3PTMN5G3HOA3BGL3UDCELAVCNFSM6AAAAACHEFJ7TKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMRSGM2DQNBVGI>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[lmoresi]

Status update — bug confirmed, tracked internally

We've confirmed the root cause: the NaturalBC namedtuple defines fn_f (lowercase) but solver code references fn_F (uppercase). Additionally, the Poisson natural BC setup has a deeper issue causing PETSc error 73 ("Object in wrong state") — the test (test_1000_poissonNaturalBC.py) is currently skipped pending a fix.

Note that the Stokes solver's natural BCs (traction conditions) work correctly — the issue is specific to the scalar Poisson solver path.

This is tracked in our internal planning and will be addressed as part of the solver cleanup work. We'll update this issue when we have a fix.

@Belay-129 — in the meantime, Dirichlet BCs work correctly for the Poisson solver. If your problem allows it, you may be able to reformulate using a Dirichlet condition on the dependent variable rather than a Neumann flux condition.

[lmoresi]

Fixed in PR #60 (merged to development)

The root cause was two bugs in the scalar Poisson solver (SNES_Scalar):

1. Natural BCs were registered with per-boundary labels ("Bottom") instead of the consolidated "UW_Boundaries" label
2. The compiled boundary residual/jacobian functions were never passed to PETSc's DS system — the code was completely missing from _setup_solver

Both issues were present in the Scalar solver but not the Vector (Stokes) solver, which is why Stokes natural BCs (traction conditions) worked correctly.

The fix mirrors the working Stokes pattern. Verified with an analytical solution (T = x²y with mixed Dirichlet/Neumann BCs) on simplex and quad meshes.

@Belay-129 — this fix is now on the development branch. Your original code pattern should work:

poisson.add_dirichlet_bc([17.0], "Top")
poisson.add_natural_bc([0.065], "Bottom")
poisson.solve()