NVIDIA/physicsnemo v2.1.0

v2.1.0

Repository: NVIDIA/physicsnemo

Tag: v2.1.0

Published: 2026-05-27T01:18:10Z

Prerelease: no

Release notes: PhysicsNeMo General Release v2.1.0

Added

• Adds GLOBE model (physicsnemo.experimental.models.globe.model.GLOBE),

including new variant that uses a dual tree traversal algorithm to reduce the complexity of the kernel evaluations from O(N^2) to O(N).

• Adds GLOBE AirFRANS example case (examples/cfd/external_aerodynamics/globe/airfrans)
• Adds GLOBE DrivAerML example case (examples/cfd/external_aerodynamics/globe/drivaer)
• Adds drop-test dynamics recipe.
• Adds concrete dropout uncertainty quantification for GeoTransolver. Learnable

per-layer dropout rates enable MC-Dropout inference for uncertainty estimates. Disabled by default (concrete_dropout: false).

• Adds automatic support for FSDP and/or ShardTensor models in checkpoint save/load

functionality

• PhysicsNeMo-Mesh now supports conversion from PyVista/VTK/VTU meshes that may

contain polyhedral cells.

• In PhysicsNeMo-Mesh, adds Mesh.to_point_cloud(), .to_edge_graph(), and

.to_dual_graph() methods. These allow Mesh conversion to 0D point clouds, 1D edge graphs, and 1D dual graphs, respectively, when connectivity information is not needed.

• Adds physicsnemo.mesh.generate subpackage with marching_cubes for

isosurface extraction from 3D scalar fields, returning a Mesh object. Supports the NVIDIA Warp backend.

• Adds a type system to PhysicsNeMo-Mesh, allowing annotation of Mesh dimensions

using notation like Mesh[2, 3] for a 2D manifold in 3D space.

• Adds adjacency caching to PhysicsNeMo-Mesh Mesh objects, allowing efficient

reuse of neighbor information.

• Adds DomainMesh class for grouping an interior mesh with named boundary

meshes and domain-level metadata, with passthrough geometric transforms (translate, rotate, scale, transform) and data operations.

• Allows selective per-field transformation of Mesh objects: transform_point_data,

transform_cell_data, and transform_global_data now accept bool | TensorDict (or plain dict for convenience).

• Adds physicsnemo.mesh.remeshing subpackage with partition_cells() for

creating Voronoi regions around seed points. BVH-accelerated.

• Added support for 1D, 2D, and 3D neighborhood attention (natten) via

physicsnemo.nn.functional interface, with full ShardTensor support.

• Added derivative functionals in physicsnemo.nn.functional for

uniform_grid_gradient, rectilinear_grid_gradient, spectral_grid_gradient, meshless_fd_derivatives, mesh_lsq_gradient, and mesh_green_gauss_gradient.

• Adds physicsnemo.sym module for symbolic PDE residual computation

(PhysicsInformer). Users define PDEs via SymPy and select a gradient method (autodiff, finite_difference, spectral, meshless_finite_difference, least_squares); spatial derivatives are computed automatically using the nn.functional.derivatives functionals.

• Ports all physics-informed examples (LDC PINNs, Darcy, Stokes MGN, DoMINO,

datacenter, xaeronet, MHD/SWE PINO) to the new physicsnemo.sym interface, replacing the separate physicsnemo-sym package dependency. Geometry is now handled via physicsnemo.mesh and PyVista.

• Added geometry functionals in physicsnemo.nn.functional for

mesh_poisson_disk_sample, mesh_to_voxel_fraction, and signed_distance_field.

• Adds embedded OOD guardrail OODGuard at

physicsnemo.experimental.guardrails.embedded, optionally wired into GeoTransolver via a new guard_config constructor argument. The guard calibrates per-channel global bounds and a geometry-latent kNN threshold during training, and emits warnings on out-of-distribution inputs at inference.

• In PhysicsNeMo-Mesh, physicsnemo.mesh.geometry now publicly exposes

stable_angle_between_vectors and compute_triangle_angles (previously only available via the private physicsnemo.mesh.curvature._utils).

• PhysicsNeMo Datapipes enables reproducability through torch.generator

utilities.

• PhysicsNeMo Datapipes now supports physicsnemo.mesh.Mesh and

physicsnemo.mesh.DomainMesh objects for deserialization, with transformations and utilities for mesh-based datasets.

• PhysicsNeMo Datapipes now support MultiDataset construction,

allowing on-the-fly construction of multi-source composite datasets that can be sampled and processed efficiently and coherently as one dataset.

• PhysicsNeMo Datapipes also support random augmentations for

mesh-based datapipes, leveraging torch.distributions for broad random distribution support. Mesh and DomainMesh datasets allow random translation, scaling, and rotation of mesh data in coherent ways, compatible with reproducability features of physicsnemo datapipes.

• Adds a new *unified* training recipe for external aerodynamics

that supports training on multiple datasets (DrivaerML, ShiftSUV, HighLiftAeroML, or more, bring your own, mix and match), supports training several different models (Domino, Transolver, GeoTransolver, Flare, GeoTransolver with Flare-attention, bring your own!). Leverages mesh datasets and non-dimensionalization to enable dataset mixing and matching at runtime. Train with surface or volume data.

• Adds a new physicsnemo.diffusion.multi_diffusion subpackage that

scales 2D diffusion models to large domains via patch-based training and inference. Provides MultiDiffusionModel2D (wraps a base model and handles state patching, conditioning preprocessing, positional-embedding injection, and per-patch output fusion), the MultiDiffusionMSEDSMLoss / MultiDiffusionWeightedMSEDSMLoss losses for patch-based DSM training, and MultiDiffusionPredictor for sampling (plugs straight into sample() / get_denoiser() and the standard solvers). Patching primitives (BasePatching2D, GridPatching2D, RandomPatching2D) are exposed under the same subpackage and are torch.compile-friendly with fullgraph=True. MultiDiffusionPredictor supports memory-efficient inference on large domains via chunk_size and use_checkpointing. The subpackage also ships patch-local DPS guidance: MultiDiffusionDPSScorePredictor (drop-in score predictor that plugs into the standard sampling stack), MultiDiffusionDataConsistencyDPSGuidance for inpainting and sparse data assimilation, and MultiDiffusionModelConsistencyDPSGuidance for generic patch-local observation operators. Use these instead of the global DPSScorePredictor to run guided sampling on domains that would otherwise OOM.

• Adds…