Relationship between time-instant number and precision of ZeaD formulas with proofs (Q820734): Difference between revisions

For time-varying numerical matrix problems, the paper studies the coefficients \(a_i\) of look-ahead and convergent discretization formulas \(\dot x_k = a_{m-1} x_{k+1}/\tau + \dots + a_0 x_{k-m+2}/\tau + O(\tau^p)\) that are used in Zhang Neural Networks (ZNN). The standard approach uses \(m\) Taylor expansions of \(x_{k+1}\) through \(x_{k-m+2}\) around \(x_k\) for a constant sampling gap \(\tau = x_j - x_{j-1}\). Forming the linear combination of these Taylor expansions with the coefficients \(a_{m-1}, \dots, a_0\) in turn then leads to a linear system for feasible \(a_j\) coefficients. If \(m > p+1\), then the linear system in \(a_j\) is solvable over the rational numbers and the relations between the coefficients can be exploited to reduce the solution set further. This method is applied to exhibit and study look-ahead and convergent finite difference formulas for various \(m\) and \(p\) values in light of their length \(m\) and their innate truncation error order \(O(\tau^p)\) on a time-varying quadratic optimization problem.