Recursive construction of biorthogonal polynomials for handling polynomial regression

An adaptive procedure for constructing a series of biorthogonal polynomials to a basis of monomials spanning the same finite-dimensional inner product space is proposed. By taking advantage of the orthogonality of the original basis, our procedure circumvents the well-known instability problem arising from the matrix inversion involved in classical polynomial regression. Moreover, the recurrent generation of biorthogonal polynomials in our framework facilitates the upgrading of all polynomials to include one additional element in the set whilst also allowing for a natural downgrading of the polynomial regression approximation. This is achieved by the posterior removal of any basis element leading to a straightforward approach for reducing the approximation order. We illustrate the usefulness of this approach through a series of examples where we derive the resulting biorthogonal polynomials from Legendre, Laguerre, and Chebyshev orthogonal bases.