A resultant matrix for scaled Bernstein polynomials (Q5926399): Difference between revisions

Let \(r(x)=\sum_{j=1}^n r_j(1-x)^{n-j}x^j\) and \(s(x)=\sum_{j=0}^m s_j (1-x)^{m-j}x^j\) be two polynomials expressed in terms of the Bernstein basis \(\{(1-x)^ux^v\}\). The author obtains a companion matrix \(M\) for one of these, say \(r(x)\), directly without having to transform the polynomials to standard form; thus \(r(x)\) is a constant multiple of the characteristic polynomial of \(M\). The elements of \(M\) turn out to be simple linear combinations of the coefficients \(r_i\). The determinant of the resultant matrix \(s(M)\) is equal to 0 if and only if \(r(x)\) and \(s(x)\) have a common root. Indeed, if \(w(x)\) is the greatest common divisor of \(r(x)\) and \(s(x)\), then the degree of \(w(x)\) is equal to \(n-\text{rank } s(M)\) and the coefficients of \(w(x)\) are proportional to the last row of \(s(M)\) after reduction to row echelon form. The author illustrates with an example; he shows that the minimal polynomial of \(M\) is its characteristic polynomial and that it is similar to its transpose via an upper triangular Hankel matrix. Computationally, his method of resultant computation is superior to the method that requires transformation of the polynomial to standard form.