Computational chaos - a prelude to computational instability (Q920269): Difference between revisions

The author is mainly concerned with the way in which chaotic behaviour can occur when the vector system \(x'=F(x)\) is approximated by the Euler differencing scheme \(x_{n+1}=x_ n+\tau F(x_ n).\) He points out that when there are many variables (as in weather-forecasting), it is tempting to take \(\tau\) large for economy's sake, but then the general predicted behaviour may become suspect. To indicate what can happen, he studies two 2-dimensional systems in some detail, with the aid of a computer. He is guided by the principle (for which he gives a non-rigorous proof) that if a system \(x_{n+1}=G(x_ n)\) has an attractor A, and G maps two points of A to one, then chaos will be present. His first system is: \(\dot x=x(a-y)\), \(y'=-y+x^ 2\), and is obtained from the well-known Lorenz system on \({\mathbb{R}}^ 3\) by letting one parameter tend to \(\infty\) and rescaling. The resulting Euler system has a cubic curve D which divides the plane into two regions, with G invertible on one but not on the other. Computer picture show the way in which the attractor changes from a smooth `ellipse', and the author studies the way in which small loops start to overlap in a subsidiary curve. The spirit of the paper is exemplified by the author's remark: `we feel that by examining the case \(\tau =1.785\), we have virtually caught chaos in the fact of forming'. Next he considers the system \(\dot x=x-y-x^ 3\), \(\dot y=x-x^ 2y\) of which all orbits (except 0) are attracted to the unit circle as a limit cycle. Again, the Euler system displays chaos, so he then passes to a 4th order Runge Kutta scheme which might be expected to be more accurate. Alas, no: he displays the attractor for the cases \(\tau =0.91\), 1.6, 1.7 and the first is smooth, but the author comments that `as a drawing of a circle it leaves much to be desired, while the others bear no resemblance to the circles they are supposed to approximate'. The paper contains much detail, described in a very readable manner.