The elements of A^-1 at the positions of the factor's pattern. It is what PEV and accuracy read, and the only part of the inverse one can afford: the full inverse of the coefficient matrix is dense.
Arguments
- a
list i, j, x, n with the triplets of ONE triangle of the symmetric matrix (the lower, i >= j, as
a_inverse()returns, or the upper); both triangles stop the call- block
0 detects the dense tail and uses the closed form; 1 forces the pure recurrence, which exists so the tests can compare the two paths.
Value
a list i, j, x, n with the computed elements of the inverse in triplets, on
the pattern of the factor. A position outside that pattern is absent, which is not
the same as being zero.