Solver objects¶
Solver constructors (for example spsolve(), cg()) return objects that implement the OpenSees
PythonSparse protocol. Application code configures the solver, passes
solver.to_openseespy() to OpenSeesPy, and inspects cached data after analysis.
Common methods¶
to_openseespy()
: Return a dict for OpenSeesPy; pass it unchanged. See
to_openseespy().
solve(**kwargs)
: Called by OpenSees; not normally invoked from application code.
formAp(**kwargs) (linear solvers only)
: Matrix-vector product Ap = A @ p.
Linear solver attributes¶
All cached attributes are None until the first solve (or formAp for A).
A
: scipy.sparse.spmatrix or cupyx.scipy.sparse.spmatrix — cached system
matrix from the last solve or formAp call.
b
: numpy.ndarray (CPU backends) or cupy.ndarray (GPU) — 1-D right-hand
side in the solver's compute dtype from the last solve.
x
: numpy.ndarray (CPU) or cupy.ndarray (GPU) — 1-D solution from the last
solve. On CPU backends this is the OpenSees x buffer (float64).
stats
: Runtime statistics (num_solves, last_solve_time, last_info,
last_num_iterations, last_residual_norm, last_error).
Counters and timing update only when constructed with record_stats=True.
last_residual_norm additionally requires debug=True (extra SpMV);
otherwise it remains None.
Eigen solver attributes¶
K, M
: scipy.sparse.spmatrix or cupyx.scipy.sparse.spmatrix — cached stiffness
and mass matrices from the last eigen solve, or None before the first solve.
stats
: Runtime statistics (num_solves, last_solve_time, last_num_modes,
last_eigenvalues, last_info, last_error).
copy.copy¶
copy.copy(solver) returns a new instance with the same keyword configuration
but reset internal cache and statistics. OpenSees uses this when cloning a
system of equations.
Compute precision (dtype) (experimental)¶
Set debug=True at construction to propagate exceptions from solve and
formAp instead of returning a negative status code (linear solvers only).
Set record_stats=True to update solver.stats after each linear solve.
Relative residual (last_residual_norm) is computed only when both
record_stats=True and debug=True (extra sparse matvec).
Compute precision (dtype) (experimental)¶
Every solver constructor accepts dtype (default numpy.float64). Supported values are
float32 and float64. The internal sparse solve, matrix cache (A,
K, M), and right-hand sides use that precision.