Numerical conditioning in structural solutions: A proposal for a new condition number (Q1342116)

The displacement method for analyzing mechanical framed structures leads to a linear system of the form (1) \(Kx= p\), where \(K\) is the stiffness matrix, \(p\) is the nodal forces vector and \(x\) denotes the unknown nodal displacement vector. The generation of \(K\) and \(p\) as well as the solution of (1) on the computer is accompanied by rounding errors. A rigorous analysis of these errors was given in the classical monograph by \textit{J. H. Wilkinson} [Rounding errors in algebraic processes (1963); see the German translation (1969; Zbl 0187.094)]. The number \(s\) of correct decimal significant digits in the solution can roughly be estimated by the formula (2) \(s= t'- \log(\kappa(K))\), where \(t'\) is the number of decimal digits associated with the binary representation used by the computer and \(\kappa(K)\) denotes the spectral condition number. The author discusses improvements of formula (2) and compares these estimates for the correct decimal digits numerically.