[Docs] Add user-guide page for matrix decompositions and solvers#643

Merged

hughperkins merged 26 commits intomainfrom

hp/docs-linalg-decompositions

May 7, 2026

Commits on May 7, 2026

[Docs] Add user-guide page for matrix decompositions and solvers
hughperkins
committed
[Docs] decompositions: drop qipc references
hughperkins
committed
[Docs] decompositions: replace ambiguous 'size' with 'dimension' (the matrix dim n)
hughperkins
committed
[Docs] decompositions: drop redundant 'fixed-size' qualifier in intro
hughperkins
committed
[Docs] decompositions: switch to unambiguous 'shape' (e.g. 2x2, 3x3) terminology
hughperkins
committed
[Docs] decompositions: FIXME note on misleading '2D/3D matrices' wording
hughperkins
committed
[Docs] decompositions: drop 'tracked separately' from FIXME note
hughperkins
committed
[Docs] decompositions: drop chatty 'be loud about it' aside
hughperkins
committed
[Docs] decompositions: state SVD ordering precisely (2x2 sorted, 3x3 not)
hughperkins
committed
[Docs] decompositions: FIXME on SVD sort-consistency 2x2 vs 3x3
hughperkins
committed
[Docs] decompositions: mention boolean template option in sort-consistency FIXME
hughperkins
committed
[Docs] decompositions: FIXME on sym_eig sort consistency (also vs svd direction)
hughperkins
committed
[Docs] decompositions: qualify qd.solve singular-A assert as debug-only
hughperkins
committed
[Docs] decompositions: split qd.solve usage bullet into three clearer ones
hughperkins
committed
[Docs] decompositions: drop LU-reuse and launch-overhead asides for qd.solve
hughperkins
committed
[Docs] decompositions: align ARAP commentary with code form (R.determinant() < 0.0)
hughperkins
committed
[Docs] decompositions: drop dangling 'one-off' qualifier on qd.solve
hughperkins
committed
[Docs] decompositions: drop IPC references
hughperkins
committed
[Docs] decompositions: drop graphics/soft-body aside in conditioning bullet
hughperkins
committed
[Docs] decompositions: drop trailing 'standard ARAP trick' aside
hughperkins
committed
[Docs] decompositions: trim conditioning bullet to f32-default statement
hughperkins
committed
[Docs] decompositions: remove 'Shape cap is the dominant constraint' bullet
hughperkins
committed
[Docs] decompositions: remove 'Numerical conditioning' bullet
hughperkins
committed
[Docs] decompositions: drop overconfident bit-exactness aside
hughperkins
committed
[Docs] decompositions: soften 'would not be' to 'may not be' for compile-time claim
hughperkins
committed
[Docs] decompositions: remove 'What's missing' section
hughperkins
committed