`scipy.linalg._flapack:csysvx`

Summary

Solve a symmetric indefinite system with condition estimation and error bounds (LAPACK csysvx).

a : ndarray: Symmetric matrix of shape (n, n).
b : ndarray: Right-hand side(s) of shape (n, nrhs).
af : ndarray, optional: Factorization to reuse when factored=1; otherwise it is computed.
ipiv : ndarray, optional: Pivot indices to reuse when factored=1, 1-based; otherwise computed.
lwork : int, optional: Size of the workspace. Default is max(3 * n, 1); the minimum is 3 * n for the real flavors and 2 * n for the complex ones. Use sysvx_lwork for the optimal value.
factored : int, optional: If nonzero, af and ipiv already hold the factorization and are reused. An integer here rather than the 'F'/'N' letter gesvx takes. Default is 0.
lower : int, optional: If nonzero, the lower triangle of a is referenced; otherwise the upper. Default is 0.
overwrite_a : int, optional: If nonzero, a may be overwritten in place. Default is 0.
overwrite_b : int, optional: If nonzero, b may be overwritten in place. Default is 0.

a_s : ndarray: a, unchanged; returned so the caller can chain calls.
udut : ndarray: The block diagonal D and the multipliers.
ipiv : ndarray: Pivot indices describing the interchanges and the 1x1 / 2x2 block structure, 1-based. Pass them on to sytrs/sytri/sycon unchanged.
b_s : ndarray: b, unchanged.
x : ndarray: Solution of the system.
rcond : float: Estimate of the reciprocal condition number.
ferr : ndarray: Estimated forward error bound for each solution vector.
berr : ndarray: Componentwise relative backward error of each solution vector.
info : int: 0 on success; if negative, the -info-th argument had an illegal value; if 0 < info <= n, d[info-1, info-1] is exactly zero; if info = n+1, the matrix is singular to working precision.