structsolve 0.6.0__tar.gz → 0.6.1__tar.gz

{structsolve-0.6.0/structsolve.egg-info → structsolve-0.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: structsolve
-Version: 0.6.0
+Version: 0.6.1
 Summary: Structural analysis solvers
 Author-email: "Saullo G. P. Castro" <castrosaullo@gmail.com>
 License: BSD-2-Clause
@@ -21,6 +21,7 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
 Classifier: Operating System :: Microsoft :: Windows
 Classifier: Operating System :: Unix
+Classifier: Environment :: WebAssembly :: Emscripten
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
@@ -62,7 +63,7 @@ Currently these solvers are pretty much compatible with my other repositories
 Citing this library
 ===================
 
-Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.0). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
+Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.1). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
 
 
 Documentation
@@ -77,9 +78,10 @@ History
 
 See [CHANGELOG.md](CHANGELOG.md) for the details of each version.
 
-* version 0.6.0 (2026-10-08)
+* version 0.6.1 (2026-10-08)
+    - Pyodide support: `await micropip.install("structsolve")` in the browser, and the whole test suite runs in Pyodide in GitHub Actions
     - `lb` and `freq` always run the sparse solvers first, whatever the size of the problem, with verified results; the dense solvers are only a last resort within `max_dense_size`, with a `DenseFallbackWarning`
-    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`)
+    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`), computing more than 15 eigenpairs in passes whose eigenvectors are made orthonormal with a Rayleigh-Ritz projection
     - Unsymmetric `lb`: a shift search finds the lowest positive real load multiplier, instead of silently returning negative or complex ones
     - The test extra no longer requires `buckling` and `composites`
 * version 0.5.3 (2026-10-06)

{structsolve-0.6.0 → structsolve-0.6.1}/README.md

@@ -27,7 +27,7 @@ Currently these solvers are pretty much compatible with my other repositories
 Citing this library
 ===================
 
-Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.0). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
+Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.1). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
 
 
 Documentation
@@ -42,9 +42,10 @@ History
 
 See [CHANGELOG.md](CHANGELOG.md) for the details of each version.
 
-* version 0.6.0 (2026-10-08)
+* version 0.6.1 (2026-10-08)
+    - Pyodide support: `await micropip.install("structsolve")` in the browser, and the whole test suite runs in Pyodide in GitHub Actions
     - `lb` and `freq` always run the sparse solvers first, whatever the size of the problem, with verified results; the dense solvers are only a last resort within `max_dense_size`, with a `DenseFallbackWarning`
-    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`)
+    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`), computing more than 15 eigenpairs in passes whose eigenvectors are made orthonormal with a Rayleigh-Ritz projection
     - Unsymmetric `lb`: a shift search finds the lowest positive real load multiplier, instead of silently returning negative or complex ones
     - The test extra no longer requires `buckling` and `composites`
 * version 0.5.3 (2026-10-06)

{structsolve-0.6.0 → structsolve-0.6.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "structsolve"
-version = "0.6.0"
+version = "0.6.1"
 description = "Structural analysis solvers"
 readme = "README.md"
 license = {text = "BSD-2-Clause"}
@@ -31,6 +31,7 @@ classifiers = [
     "Programming Language :: Python :: 3.14",
     "Operating System :: Microsoft :: Windows",
     "Operating System :: Unix",
+    "Environment :: WebAssembly :: Emscripten",
 ]
 dependencies = [
     "numpy",

{structsolve-0.6.0 → structsolve-0.6.1}/structsolve/arpackutils.py

@@ -11,7 +11,9 @@ MKL 2025.0.1 or newer is not affected, neither are the non-symmetric drivers
 of :func:`scipy.sparse.linalg.eigs`, which do not call ``dsteqr``.
 
 :func:`capped_eigsh` keeps ``ncv <= 32`` when :func:`arpack_max_ncv` says so,
-computing more than 15 eigenpairs in several calls with deflation.
+computing more than 15 eigenpairs in several calls with deflation, whose
+eigenvectors are combined with a Rayleigh-Ritz projection such that they
+are orthonormal also across the calls.
 
 The cap is decided by :func:`arpack_max_ncv` and can be overridden with
 the module attribute :data:`ARPACK_MAX_NCV` or the environment variable
@@ -32,6 +34,7 @@ import sys
 import warnings
 
 import numpy as np
+from scipy.linalg import cholesky, eigh, solve_triangular
 from scipy.sparse import csc_matrix
 from scipy.sparse.linalg import LinearOperator, splu
 from scipy.sparse.linalg import eigsh as _scipy_eigsh
@@ -281,6 +284,34 @@ def _deflation_shift(w, far_shift):
     return 0.5*(last[i] + last[i + 1])
 
 
+def _rayleigh_ritz(A, M, V):
+    r"""Rayleigh-Ritz projection of `[A]\{x\} = w [M]\{x\}` onto `[V]`
+
+    The eigenvectors of different passes of :func:`capped_eigsh` are only
+    `[M]`-orthogonal to the accuracy of ARPACK, e.g. `10^{-6}` between
+    close eigenvalues of different passes. `[V]` is `[M]`-orthonormalized
+    with the Cholesky factor `[L]` of `[V]^T[M][V]`, and the projected
+    problem `[W]^T[A][W]\{y\} = w\{y\}`, with `[W] = [V][L]^{-T}`, is
+    solved with the LAPACK driver ``dsyevr`` (``driver='evr'``), which does
+    not use the ``dsteqr`` of Intel MKL broken above 32 x 32 matrices, see
+    :func:`dsteqr_is_faulty`. The returned eigenvectors `[W]\{y\}` are
+    `[M]`-orthonormal to the rounding errors, and the Ritz values are at
+    least as accurate as the eigenvalues of the passes.
+
+    Returns
+    -------
+    w, x : ndarray
+        The Ritz values by increasing value and the Ritz vectors.
+
+    """
+    MV = M @ V
+    L = cholesky(0.5*(V.T @ MV + MV.T @ V), lower=True)
+    W = solve_triangular(L, V.T, lower=True).T
+    H = W.T @ (A @ W)
+    w, Y = eigh(0.5*(H + H.T), driver='evr')
+    return w, W @ Y
+
+
 def capped_eigsh(A, k, M, sigma, which, mode='normal', tol=0):
     r""":func:`scipy.sparse.linalg.eigsh` with ``ncv`` from :func:`arpack_ncv`
 
@@ -312,7 +343,9 @@ def capped_eigsh(A, k, M, sigma, which, mode='normal', tol=0):
     in the study of the panels cylinders it was 30 to 45% faster and gave
     eigenvalues 5 to 10 times more accurate. Each pass factorizes
     `[A] - \sigma_d [M]` with SuperLU, and every call starts from
-    :func:`start_vector`.
+    :func:`start_vector`. The eigenvectors of all the passes are finally
+    combined with a Rayleigh-Ritz projection, see :func:`_rayleigh_ritz`,
+    such that they are `[M]`-orthonormal also across the passes.
 
     Returns
     -------
@@ -359,5 +392,4 @@ def capped_eigsh(A, k, M, sigma, which, mode='normal', tol=0):
         w = np.concatenate((w, wi))
         V = np.hstack((V, Vi))
         release_memory()
-    order = np.argsort(w, kind='stable')
-    return w[order], V[:, order]
+    return _rayleigh_ritz(A, M, V)

{structsolve-0.6.0 → structsolve-0.6.1}/structsolve/linear_buckling.py

@@ -862,9 +862,10 @@ def lb(K, KG, tol=0, sparse_solver=True, silent=False,
     MKL versions returns wrong eigenvectors for matrices larger than
     32 x 32, which ARPACK uses with the size ``ncv``. When such a LAPACK is
     detected, ``ncv`` is capped at 32 and more than 15 eigenpairs are
-    computed in several passes, see :mod:`structsolve.arpackutils`, where
-    the cap can be overridden, e.g. with the environment variable
-    ``STRUCTSOLVE_ARPACK_MAX_NCV=0`` to lift it with MKL 2025.0.1 or newer.
+    computed in several passes, combined with a Rayleigh-Ritz projection,
+    see :mod:`structsolve.arpackutils`, where the cap can be overridden,
+    e.g. with the environment variable ``STRUCTSOLVE_ARPACK_MAX_NCV=0`` to
+    lift it with MKL 2025.0.1 or newer.
     The non-symmetric drivers of :func:`scipy.sparse.linalg.eigs` are not
     affected.
 

{structsolve-0.6.0 → structsolve-0.6.1/structsolve.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: structsolve
-Version: 0.6.0
+Version: 0.6.1
 Summary: Structural analysis solvers
 Author-email: "Saullo G. P. Castro" <castrosaullo@gmail.com>
 License: BSD-2-Clause
@@ -21,6 +21,7 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
 Classifier: Operating System :: Microsoft :: Windows
 Classifier: Operating System :: Unix
+Classifier: Environment :: WebAssembly :: Emscripten
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 License-File: LICENSE
@@ -62,7 +63,7 @@ Currently these solvers are pretty much compatible with my other repositories
 Citing this library
 ===================
 
-Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.0). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
+Saullo G. P. Castro (2026). Structural analysis solvers tailored for semi-analytical models (Version 0.6.1). Zenodo. DOI: https://doi.org/10.5281/zenodo.2581212.
 
 
 Documentation
@@ -77,9 +78,10 @@ History
 
 See [CHANGELOG.md](CHANGELOG.md) for the details of each version.
 
-* version 0.6.0 (2026-10-08)
+* version 0.6.1 (2026-10-08)
+    - Pyodide support: `await micropip.install("structsolve")` in the browser, and the whole test suite runs in Pyodide in GitHub Actions
     - `lb` and `freq` always run the sparse solvers first, whatever the size of the problem, with verified results; the dense solvers are only a last resort within `max_dense_size`, with a `DenseFallbackWarning`
-    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`)
+    - ARPACK's `ncv` is capped at 32 when SciPy is linked against the Intel MKL versions whose `dsteqr` is faulty (new `structsolve.arpackutils`), computing more than 15 eigenpairs in passes whose eigenvectors are made orthonormal with a Rayleigh-Ritz projection
     - Unsymmetric `lb`: a shift search finds the lowest positive real load multiplier, instead of silently returning negative or complex ones
     - The test extra no longer requires `buckling` and `composites`
 * version 0.5.3 (2026-10-06)

{structsolve-0.6.0 → structsolve-0.6.1}/tests/test_linear_buckling_unsymmetric.py

@@ -217,6 +217,9 @@ def test_lb_nearly_symmetric(sparse_solver, max_dense_size):
     assert np.all(relative_residuals(Kn, KG, eigvals, eigvecs) < 1e-8)
 
 
+@pytest.mark.skipif(sys.platform == 'emscripten',
+                    reason='no subprocesses in Pyodide, whose SciPy is not '
+                           'linked to MKL')
 def test_eig_complex_mkl_crash_regression():
     """Regression test: with Intel MKL 2025.0.0, scipy.linalg.eig on these
     real 48 x 48 matrices (LAPACK dggev) corrupts the heap and the process