Compute sgn, scale, M such that $$P = sgn * scale * dot(M, t(M))$$.

.decomp_quad(P, cond = NA, rcond = NA)

## Arguments

P

A real symmetric positive or negative (semi)definite input matrix

cond

Cutoff for small eigenvalues. Singular values smaller than rcond * largest_eigenvalue are considered negligible.

rcond

Cutoff for small eigenvalues. Singular values smaller than rcond * largest_eigenvalue are considered negligible.

## Value

A list consisting of induced matrix 2-norm of P and a rectangular matrix such that P = scale * (dot(M1, t(M1)) - dot(M2, t(M2)))