Add an independent AB-UPT operator and unified aero recipe
Maintainers usually reply within 2 days
@mnabian is already working on this.
Since Oct 23, 2026.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 25/100
- Issue type
- Feature
- Clarity
- Mostly clear
- Activity status
- Active
- Domain
- ai, machine-learning
Research direction
Start by reading arXiv:2502.09692v4 and inspect the existing PhysicsNeMo components that support the requested model and aerodynamics recipe. The scope includes model implementation, training modes, sampling, checkpoint round trips, and physical-field export; done means these areas are covered without claiming numerical parity or benchmark reproduction.
Written by the indexing model from the issue text.
Description
Implement the Anchored-Branched Universal Physics Transformer from arXiv:2502.09692v4, using the paper and existing PhysicsNeMo components without research implementation source code or weights.
Scope:
- Experimental operator with geometry supernode pooling, interacting surface/volume anchors, reusable query contexts, and a physically scaled finite-difference vorticity helper.
- Separate model and unified external aerodynamics recipe PRs.
- Surface-only, volume-only, and joint training with unequal query counts, reproducible sampling, checkpoint round trips, and physical-field export.
This is an independent implementation, not a claim of numerical parity or reproduction of benchmark accuracy.
- Dominant language
- Python
- Stars
- 3.4k
- Forks
- 828
- Avg merge
- 4d 4h
- Merged PRs (30d)
- 78
Getting set up
- Ships a Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA/physicsnemo
-
Wasserstein CDF distance uses the first width for every binPossibly taken @sylvesterkaczmarek claimed this 28 days ago. Open
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
NVIDIA/physicsnemo#2042 ·
Maintainers usually reply within 2 days
-
🐛[BUG]: Darcy2D Jacobi kernel divides grad(k)·grad(u) by 2dx instead of 4dx², so the data doesn't solve the Darcy equationPossibly taken @coreyjadams claimed this 26 days ago. Open
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
NVIDIA/physicsnemo#2036 · 2 comments · 1 assignee ·
Maintainers usually reply within 2 days
-
🐛[BUG]: MHD_PDE equation "B_grad_u" uses v * Bx.diff(y) instead of By * u.diff(y)Possibly taken @aaguadob claimed this 33 days ago. Open? - Needs Triage bug
Difficulty 1/5 Under an hour Newbie friendliness 92/100
NVIDIA/physicsnemo#2021 ·
Maintainers usually reply within 2 days
-
🐛[BUG]: Diffusion(time=True) in darcy_physics_informed/utils.py silently drops the time-derivative termPossibly taken @aaguadob claimed this 33 days ago. Open? - Needs Triage bug
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
NVIDIA/physicsnemo#2020 ·
Maintainers usually reply within 2 days
-
Difficulty 3/5 1-2 days Newbie friendliness 72/100
NVIDIA/physicsnemo#2207 ·
Maintainers usually reply within 2 days
All issues in NVIDIA/physicsnemo
Similar issues
-
Status: Triage Type: Quest
Difficulty 2/5 1-3 hours Newbie friendliness 66/100
GTNewHorizons/GT-New-Horizons-Modpack#27490 ·
Maintainers usually reply within 1 day
-
Bug
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
retentioneering/retentioneering-tools#231 ·
Maintainers usually reply within 1 day
-
backend bug documentation
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
geolens-io/geolens#3568 ·
Maintainers usually reply within 1 day
-
enhancement
Difficulty 1/5 Under an hour Newbie friendliness 88/100
mprov-ng/mprov_control_center#450 ·
Maintainers usually reply within 1 day