The equal coefficients quadrature rules and their numerical improvement (Q2491027)

This paper deals with new equal coefficients quadrature rules of the following kind: \[ \int_a^b w(x)f(x)\text{ d}x\simeq C_n\sum_{i=1}^n f(x_i). \tag{1} \] In particular, after presenting some examples and showing that no integration formula of Newton-Cotes type with equal coefficients (other than the trapezoidal one) exists, the authors prove that the precision degree of constant coefficients formulas based on \(n\) quadrature nodes can be increased from \(n\) to \(n+2\) by taking the upper and the lower bounds of the integration formula as unknowns. In order to do this, starting from the assumption that (1) is exact for \(f(x)=x^j,\;j=0,\dots,n\), a matrix proof is provided to show that the nonlinear system for the basis \(f(x)=x^j,\;j=0,\dots,n+2\) has no analytic solution. So the system of order \((n+2)\times(n+2)\) is numerically solved for \(x_1,\ldots,x_n\) and the two additional unknowns \(a\) and \(b\), since \(C_n={1\over n}\int_a^b w(x)\text{ d}x\) by putting \(f(x)\equiv1\) in (1). Therefore the precision degree has been approximately increased by two degrees. A great deal of numerical examples show the computational efficiency of the improved equal coefficients rules with respect to well known methods, such as the two-point trapezoidal and the \(n\)-point Gauss-Chebyshev of the first kind rules. In particular it is worthy to point out that the improved 6-point and 7-point formulas can be compared in average with the 562-point and the 408-point formulas of usual Gauss-Legendre quadrature rules for all the examples presented.