Add Von Mises Metric for Structural Mechanics Applications

[mnabian]

PhysicsNeMo Pull Request

Description

Add von Mises stress metric for structural mechanics. This will be used in the structural mechanics recipes.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• The CHANGELOG.md is up to date with these changes.
• An issue is linked to this pull request.
• If I am implementing a new model or modifying any existing model, I have followed the Models Implementation Coding Standards.

Dependencies

Review Process

All PRs are reviewed by the PhysicsNeMo team before merging.

Depending on which files are changed, GitHub may automatically assign a maintainer for review.

We are also testing AI-based code review tools (e.g., Greptile), which may add automated comments with a confidence score.
This score reflects the AI’s assessment of merge readiness and is not a qualitative judgment of your work, nor is
it an indication that the PR will be accepted / rejected.

AI-generated feedback should be reviewed critically for usefulness.
You are not required to respond to every AI comment, but they are intended to help both authors and reviewers.
Please react to Greptile comments with 👍 or 👎 to provide feedback on their accuracy.

[mnabian]

/blossom-ci

[greptile-apps]

Greptile Summary

This PR adds a von_mises_from_stress_tensor function to physicsnemo/metrics/cae/von_mises.py that computes the von Mises equivalent stress from a symmetric 3×3 (or batched N×3×3) stress tensor, supporting both PyTorch and NumPy inputs. The implementation is mathematically correct, gradient-friendly (via torch.clamp instead of raw sqrt), and is paired with a thorough test suite.

• Correct formula — the distortional-energy (J₂) formula is properly implemented and verified for uniaxial, hydrostatic, and pure-shear reference cases.
• Unused parameter — _validate_stress_shape accepts an is_torch: bool argument that is never read inside the function; it is dead code and should be removed.
• Validation order inconsistency — for the numpy path, np.asarray(stress, dtype=np.float64) is called before shape validation, which may perform an unnecessary allocation for invalid inputs; the torch path validates before computing anything.
• No 2D plane-stress support — the function is limited to full 3×3 tensors; Voigt 6-vector or 2×2 plane-stress inputs are silently rejected with a ValueError. This is acceptable for the stated scope but worth noting in the docstring.
• Changelog and tests are both present and appropriate.

Important Files Changed

Filename	Overview
physicsnemo/metrics/cae/von_mises.py	New module implementing von Mises stress from 3×3 or N×3×3 stress tensors for both PyTorch and NumPy inputs. The math is correct and the gradient path is preserved. Two minor style issues: is_torch parameter in _validate_stress_shape is unused dead code, and the numpy path converts to float64 before the shape validation runs (inconsistent with the torch path).
test/metrics/test_metrics_von_mises.py	Comprehensive test suite covering uniaxial tension, hydrostatic stress, pure shear, batched inputs (both torch and numpy), invalid shapes, autograd support, and asymmetric shear symmetrization. Correctly uses the device fixture from the project-wide conftest.py.
CHANGELOG.md	One-line addition to the Unreleased section documenting the new von Mises metric module.

_{Last reviewed commit: 42c8268}

[mnabian]

/blossom-ci

[mnabian]

/blossom-ci

[peterdsharpe]

A few changes requested, but overall this'll be a nice new feature when merged.

[peterdsharpe]

In new code, we should always prefer modern | union types over Union. (Both here and throughout)

[peterdsharpe]

These should all use jaxtyping to be compliant with repo rules. (Both here and throughout, on all type signatures for tensors)

[peterdsharpe]

For readability, and for better error handling, I'd recommend always prefixing switch-statement clauses with an:

elif isinstance(stress, np.ndarray):

And then having an:

else:
    raise TypeError("...")

In other words, switch statements should always declare their switches.

[peterdsharpe]

Missing type signature here

[peterdsharpe]

This can be vectorized, if desired, using einsum. This may yield performance benefits if this is used downstream in performance-critical applications - up to you.

It's also possible to write this code in a way that's dimensionally-generic (e.g., works on 2x2 stress tensors for 2D problems), which could be nice down the road for reusability?

[peterdsharpe]

It might be worth using the Array API standard to deduplicate the parallel numpy and torch code here? This is the official effort coordinated across NumPy/Torch/JAX/others to allow for multiple dispatch on arrays.

https://data-apis.org/array-api-compat/

I've used this in other projects - it's excellent and makes for very clean reusable implementations.

[peterdsharpe]

Suggested change

	- Adds von Mises stress (`physicsnemo.metrics.cae.von_mises`).
	- Adds von Mises stress (`physicsnemo.metrics.cae.von_mises`) as a CAE metric.