On the preservation of second integrals by Runge-Kutta methods (Q6111237)

The author uses the recently discovered theory of ``discrete second integrals'' to show novel results about Runge-Kutta methods. In particular, to prove interesting new results about such methods in a general setting. He considers an autonomous ODE \(\dot{x}=f(x)\) in \(\mathbb{R}^{n}\) and a second integral of this ODE, namely a function \(p(x)\) that satisfies \(\dot{p}(x)=f(x)^{T}\nabla p(x)=c(x) p(x)\), where \(c(x)\) is called the cofactor of \(p(x)\). The discrete-time analogue of \(p(x)\), is referred to as a ``discrete second integral'' of a map \(\varphi_{h} : \mathbb{R}^{n} \rightarrow \mathbb{R}^{n}\), which is a function \(p(x)\) that satisfies \(p(\varphi_{h}(x)) = \tilde{c}(x)p(x)\), where \(\tilde{c}(x)\) is called the discrete cofactor of \(p(x)\). Firstly, the author shows that Runge-Kutta methods cannot preserve irreducible quadratic second integrals of an ODE. Next he proves that Runge-Kutta methods always preserve affine second integrals and describe the implications of this result. He provides the following example. Consider the ODE in two dimensions \(\dot{x}= x^2 + 2 x y + 3 y^2\), \(\dot{y} = 2 y (2 x + y)\). This system of ODEs has the three linear second integrals \(p_{1}(x) = x + y\), \(p_{2}(x) = x - y\), \(p_{3}(x) = y\), that correspond to the cofactors \(c_{1}(x) = x + 5y\), \(c_{2}(x) = x - y\), \(c_{3}(x) = 4x + 2y\). The author mentions some interesting results as corollaries such as the preservation of certain rational integrals by Runge-Kutta methods. For completeness, he further provides two examples, the first one being a Lotka-Volterra system, the second a nonpolynomial vector field. Moreover, the special case of affine second integrals with constant cofactor is discussed, as well the preservation of third and higher integrals. Several examples are presented. The author also shows that for all linear ODEs with quadratic first integrals, all diagonal Padé Runge-Kutta maps preserve the integral exactly. At the end of the paper, he gives an algorithm that detects the existence of such affine integrals as well as the linear transformation that decouples the affine subsystem of higher integrals from the nonlinear components.