BUG: Cauchy-Born get_shift_gradients (F-rotation) + SinclairCrack update_precon (region order)#320
Open
jameskermode wants to merge 5 commits into
Open
BUG: Cauchy-Born get_shift_gradients (F-rotation) + SinclairCrack update_precon (region order)#320jameskermode wants to merge 5 commits into
jameskermode wants to merge 5 commits into
Conversation
…rCrack.update_precon (region order)
get_shift_gradients applied only A^T (dchi/dde), dropping the F-induced rotation R (from the polar
decomposition of the deformation gradient) and dR/dde that predict_shifts applies as s = A^T R chi.
This is exact only at R=I; near a crack tip (large F) the analytic shift gradient drifts from the true
derivative, growing with load, which breaks the configurational-force identity f_alpha = -dE/dalpha for a
flexible-BC crack with a multilattice (Cauchy-Born) corrector. Fix differentiates the full applied shift
d(A^T R chi)/dde = A^T((dR/dde)chi + R dchi/dde), dchi/dde analytic, dR/dde by the same finite difference
already taken for the strain field. New regression test test_regression_model_gradient_F_with_rotation
(rotation-bearing F, R!=I) confirms analytic==FD to ~1e-9; the existing gradient tests use symmetric F
(R=I) and could not catch this.
SinclairCrack.update_precon selected regions I+II via self.atoms[:self.N2], assuming the cluster is sorted
region I,II,III. set_regions' default sort ('r_theta_z') is not region-grouped, so the slice is not regions
I+II and the region-I preconditioner block is empty/misaligned (spilu size mismatch). Select via the boolean
mask self.regionI_II (array order), matching how update_atoms/get_forces index the DOFs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Member
Author
|
@Fraser-Birks please can you take a look at this when you have a chance? |
This was referenced Jun 30, 2026
scipy removed the deprecated disp= keyword from scipy.linalg.sqrtm; disp=True was the default (return just the matrix), so sqrtm(Usqr) is equivalent and portable across scipy versions. Fixes the 10 TestPredictCauchyBornShifts failures (find_grad / fit_taylor). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CubicCrystalDislocation.view_cyl (via _plot_DXA_disloc_line) builds a
colour key and segment name with fractions.Fraction(elem) over the DXA
Burgers vector components. Newer OVITO returns segment.true_burgers_vector
as float32, and Fraction() rejects numpy.float32 (it is not a subclass of
Python float, unlike numpy.float64) with:
TypeError: argument should be a string or a Rational instance
This broke the documentation build (docs/applications/cylinder_configurations.ipynb).
Convert each component to a native Python float before passing to Fraction.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
atomman.defect.differential_displacement still calls matplotlib.cm.get_cmap, removed in matplotlib>=3.9 (upstream atomman master is not yet compatible). Skip the test in that case rather than fail; nothing to fix matscipy-side. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
test_regression_model_gradient_F_with_rotation's model_prediction precondition used atol=5e-4, but the regression CB model's intrinsic accuracy floor for the rotated, 1%-strained shift is ~5.3e-4 (converges there as samples grow; not noise). This test never actually ran in CI before (the suite died earlier at the removed scipy sqrtm(disp=) kwarg), so the optimistic dev-machine tolerance went unnoticed. Relax the precondition to 1e-3; the actual F-rotation correctness is the analytic-vs-FD gradient check (atol=1e-7), which passes at ~1.6e-10. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two correctness bugs in the multilattice (Cauchy–Born) flexible-boundary-condition crack machinery, found while building a Si(111) Sinclair-crack reference with a Stillinger–Weber multilattice.
1.
CubicCauchyBorn.get_shift_gradientsdrops the F-induced rotationpredict_shiftsapplies the corrector shift ass_i = Aᵀ R_i χ_i, whereR_iis the rotation from the polar decomposition of the deformation gradientF_i. Butget_shift_gradientsreturnedds_i/dde = Aᵀ (dχ_i/dde)— dropping bothR_ianddR_i/dde. This is exact only at zero strain (R_i = I); near a crack tip (largeF) the analytic shift gradient drifts from the true derivative, growing with load, and breaks the configurational-force identityf_alpha = -dE/dalphafor a flexible-BC crack with a Cauchy–Born corrector (the error flips the sign of the tip force at highK).Fix: differentiate the full applied shift,
d(Aᵀ R χ)/dde = Aᵀ( (dR/dde) χ + R dχ/dde ),keeping
dχ/ddeanalytic (regression gradient) and takingdR/ddeby the same central difference already used for the strain field.Test: new
test_regression_model_gradient_F_with_rotationuses a rotation-bearingF(R ≠ I) and checks analyticget_shift_gradientsagainst the finite difference of the applied shift (agreement ~1e-9). The existingtest_regression_model_gradient_{E,F}use symmetricF(R = I) and so could not catch this. Before the fix the new test fails by ~3e-2; after, ~4e-10.2.
SinclairCrack.update_preconassumes a region-sorted clusterupdate_preconselected regions I+II viaa = self.atoms[:self.N2], assuming the cluster is ordered region I, II, III.set_regions' default sort ('r_theta_z') is not region-grouped, so the slice is not regions I+II and the region-I preconditioner sub-block is empty/misaligned (→spilusize mismatch). Select via the boolean maskself.regionI_II(array order), matching howupdate_atoms/get_forcesindex the DOFs.Validation
master.get_shift_gradientsvalidated against FD ofpredict_shiftsacross a range of crack loads (machine precision after fix; 3–6% and growing before).🤖 Generated with Claude Code