Sym upstream pr1

.github/CODEOWNERS

@@ -70,6 +70,9 @@ physicsnemo/core/ @coreyjadams @CharlelieLrt @ktangsali
 # Reusable neural network building blocks
 physicsnemo/nn/ @loliverhennigh
 
+# Symbolic PDE residual computation (PhysicsInformer)
+physicsnemo/sym/ @ktangsali
+
 # Functional operations (KNN, radius search, SDF)
 physicsnemo/nn/functional/ @loliverhennigh
 physicsnemo/nn/functional/knn/ @coreyjadams @peterdsharpe @loliverhennigh

.importlinter

@@ -13,7 +13,7 @@ containers=
 layers = 
    experimental
    active_learning : diffusion
-   models : datapipes : metrics : domain_parallel : optim
+   models : datapipes : metrics : domain_parallel : optim : sym
    mesh
    nn
    utils

CHANGELOG.md

@@ -44,6 +44,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   `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.
 
 ### Changed
 

Dockerfile

@@ -207,7 +207,7 @@ RUN if [ "$TARGETPLATFORM" = "linux/amd64" ] && [ "$NATTEN_AMD64_WHEEL" != "unkn
 RUN uv pip install --no-build-isolation "torch_sparse"
 
 # All pyproject extras (no dev); installs physicsnemo non-editable
-RUN cd /physicsnemo && uv pip install ".[cu13,utils-extras,mesh-extras,datapipes-extras,gnns]"
+RUN cd /physicsnemo && uv pip install ".[cu13,utils-extras,mesh-extras,datapipes-extras,gnns,sym]"
 
 # Cleanup builder stage
 RUN rm -rf /physicsnemo/

docs/api/physicsnemo.sym.rst

@@ -0,0 +1,19 @@
+PhysicsNeMo Sym
+===============
+
+Symbolic PDE residual computation for physics-informed training.
+
+.. autoclass:: physicsnemo.sym.eq.pde.PDE
+   :members:
+   :show-inheritance:
+
+.. autoclass:: physicsnemo.sym.eq.phy_informer.PhysicsInformer
+   :members:
+   :show-inheritance:
+
+.. autoclass:: physicsnemo.sym.eq.gradients.GradientCalculator
+   :members:
+
+.. autofunction:: physicsnemo.sym.eq.gradients.compute_connectivity_tensor
+
+.. autofunction:: physicsnemo.sym.eq.gradients.compute_stencil3d

physicsnemo/sym/__init__.py

@@ -0,0 +1,52 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 - 2026 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""PhysicsNeMo Sym: symbolic PDE residual computation.
+
+Example
+-------
+>>> from sympy import Symbol, Function
+>>> from physicsnemo.sym.eq.pde import PDE
+>>> from physicsnemo.sym.eq.phy_informer import PhysicsInformer
+>>>
+>>> class Poisson(PDE):
+...     def __init__(self):
+...         self.dim = 2
+...         x, y = Symbol("x"), Symbol("y")
+...         u = Function("u")(x, y)
+...         self.equations = {"poisson": u.diff(x, 2) + u.diff(y, 2)}
+...
+>>> pde = Poisson()
+>>> pi = PhysicsInformer(["poisson"], pde, grad_method="autodiff")
+>>> sorted(pi.required_inputs)
+['coordinates', 'u']
+"""
+
+from physicsnemo.sym.eq.gradients import (
+    GradientCalculator,
+    compute_connectivity_tensor,
+    compute_stencil3d,
+)
+from physicsnemo.sym.eq.pde import PDE
+from physicsnemo.sym.eq.phy_informer import PhysicsInformer
+
+__all__ = [
+    "GradientCalculator",
+    "PDE",
+    "PhysicsInformer",
+    "compute_connectivity_tensor",
+    "compute_stencil3d",
+]

physicsnemo/sym/computation.py

@@ -0,0 +1,101 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 - 2026 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Lightweight replacement for Sym's Node — wraps a callable with string-based I/O metadata."""
+
+from __future__ import annotations
+
+from physicsnemo.sym.constants import diff_str
+
+
+class Computation:
+    """A named unit in the computational graph.
+
+    Parameters
+    ----------
+    inputs : list[str]
+        Non-derivative input names (e.g. ``["u", "v", "p"]``).
+    outputs : list[str]
+        Output names produced by this computation.
+    evaluate : callable
+        A callable (typically ``torch.nn.Module``) mapping
+        ``Dict[str, Tensor] → Dict[str, Tensor]``.
+    name : str
+        Human-readable label for debugging.
+    """
+
+    def __init__(
+        self,
+        inputs: list[str],
+        outputs: list[str],
+        evaluate,
+        name: str = "Computation",
+    ):
+        if isinstance(inputs, str):
+            inputs = [inputs]
+        all_inputs = [str(x) for x in inputs]
+        self._inputs = [x for x in all_inputs if diff_str not in x]
+        self._derivatives = [x for x in all_inputs if diff_str in x]
+        self._outputs = [str(x) for x in outputs]
+        self.evaluate = evaluate
+        self._name = name
+
+    @classmethod
+    def from_sympy(cls, eq, out_name, freeze_terms=None, detach_names=None):
+        """Build a Computation from a SymPy expression."""
+        from physicsnemo.sym.utils.sympy.torch_printer import (
+            SympyToTorch,
+            _subs_derivatives,
+        )
+
+        if freeze_terms is None:
+            freeze_terms = []
+        if detach_names is None:
+            detach_names = []
+
+        sub_eq = _subs_derivatives(eq)
+        evaluate = SympyToTorch(sub_eq, out_name, freeze_terms, detach_names)
+        inputs = list(evaluate.keys)
+        outputs = [out_name]
+        return cls(inputs, outputs, evaluate, name="Sympy Computation: " + out_name)
+
+    @property
+    def name(self) -> str:
+        """Human-readable label for this computation."""
+        return self._name
+
+    @property
+    def outputs(self) -> list[str]:
+        """Output names produced by this computation."""
+        return self._outputs
+
+    @property
+    def inputs(self) -> list[str]:
+        """Non-derivative input names."""
+        return self._inputs
+
+    @property
+    def derivatives(self) -> list[str]:
+        """Derivative input names (contain ``__``)."""
+        return self._derivatives
+
+    def __str__(self) -> str:
+        return (
+            f"computation: {self.name}\n"
+            f"  inputs: {self.inputs}\n"
+            f"  derivatives: {self.derivatives}\n"
+            f"  outputs: {self.outputs}"
+        )

physicsnemo/sym/constants.py

@@ -0,0 +1,29 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 - 2026 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import torch
+
+diff_str: str = "__"
+
+
+def diff(y: str, x: str, degree: int = 1) -> str:
+    """Build a derivative name string: ``diff('u', 'x')`` → ``'u__x'``."""
+    return diff_str.join([y] + degree * [x])
+
+
+tf_dt = torch.float32
+np_dt = np.float32

physicsnemo/sym/eq/__init__.py

@@ -0,0 +1,31 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 - 2026 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from physicsnemo.sym.eq.gradients import (
+    GradientCalculator,
+    compute_connectivity_tensor,
+    compute_stencil3d,
+)
+from physicsnemo.sym.eq.pde import PDE
+from physicsnemo.sym.eq.phy_informer import PhysicsInformer
+
+__all__ = [
+    "GradientCalculator",
+    "PDE",
+    "PhysicsInformer",
+    "compute_connectivity_tensor",
+    "compute_stencil3d",
+]