A counterexample of the use of energy as a measure of computational accuracy (Q750096): Difference between revisions

The author gives a counterexample of the usual practice consisting to equate accuracy of the energy of the numerical solution with the accuracy of the numerical solution itself. To show this fact a concrete problem is set: in the (x,y)-plane a particle of unit mass is positioned at (0.5,0.0) and has initial velocity of (0.0,1.63). Its trajectory is to be determined if it is acted upon by a central, attractive force \({\mathcal F}\) whose magnitude F satisfies \(F=1/r^ 2\). The resulting conservative motion can be determined exactly by the methods of classical mechanics. The dynamical formulation of motion leads to two differential equations (of second order). This system is numerically solved by using an implicit scheme and Newton's method. At each time the energy is \(E_ k=-0.67155\) which coincides with the exact energy E. But numerical results confirm that accuracy in \(E_ k\) is not equivalent to accuracy of the numerical calculations. The author gives the reason of this mishap: The energy \(E_ k\) is independent of \(\Delta\) t and is always the same as E.