Release Notes

0.8.0

Highlights

Complete Matrix-Vector Calculus — Solverz now fully supports symbolic differentiation of mixed matrix-vector equations. Write equations like e*(G@e - B@f) + f*(B@e + G@f) - P and get analytical Jacobians automatically.

Unified Mat_Mul Interface — Mat_Mul(A, x) replaces the legacy MatVecMul as the standard matrix-vector product. It uses scipy.sparse directly (faster than the old csc_matvec) and supports full matrix calculus.

New Features

  • Matrix calculus operators: exp, sin, cos, ln, power (**), transpose, and Diag now work inside matrix-vector expressions with automatic differentiation.

    from Solverz import Mat_Mul, Var, Param, Model, Eqn
from Solverz.sym_algebra.functions import exp

m = Model()
m.A = Param('A', [[1, 0], [0, 2]], dim=2, sparse=True)
m.x = Var('x', [1, 1])
m.f = Eqn('f', exp(Mat_Mul(m.A, m.x)))  # Jacobian: diag(exp(A@x)) @ A

  • Mutable matrix Jacobian: Variable-dependent matrix derivatives (e.g., diag(e)@G + diag(f)@B from power flow equations) are now evaluated dynamically at each Newton step. The sparsity pattern is determined at initialization and remains fixed; only the data values are updated.

  • Selective Numba @njit: In module mode, Numba compilation is applied selectively — equations using Mat_Mul run with scipy.sparse (fast C-level sparse operations), while pure element-wise equations retain @njit acceleration.

  • atan2 symbolic function: Added atan2(y, x) for computing the two-argument arctangent in symbolic equations.

  • Plugin-based module discovery: Third-party numerical modules (e.g., SolMuseum) are now discovered via entry_points(group='solverz.num_api') instead of hard-coded imports. Packages register via pyproject.toml [project.entry-points."solverz.num_api"]. Closes #118.

  • Improved solution stats: Solvers now record more detailed statistics and profiling information in the solution object.

Bug Fixes

  • Stabilized solution slicing and incidence matrix helpers.

Deprecations

  • MatVecMul is deprecated — use Mat_Mul(A, x) instead. MatVecMul will emit a DeprecationWarning when used. It will be removed in a future release.

    # Before (deprecated):
from Solverz import MatVecMul
m.f = Eqn('f', MatVecMul(m.A, m.x) - m.b)

# After (recommended):
from Solverz import Mat_Mul
m.f = Eqn('f', Mat_Mul(m.A, m.x) - m.b)

Documentation

  • New: Matrix-Vector Calculus — functionality, mathematical background, and application examples (power flow, heat network, nonlinear equations).

  • New: Extending Matrix Calculus — developer guide for adding new operations to the matrix calculus module.

  • Updated: Getting Started — matrix equation examples now use Mat_Mul.

Full Changelog

0.7.2…0.8.0