Professional Documents
Culture Documents
A Modified Runge-Kutta Method For Nonlinear Dynamical Systems With Conserved Quantities PDF
A Modified Runge-Kutta Method For Nonlinear Dynamical Systems With Conserved Quantities PDF
1 Motivation system (1) to obtain a numerical solution, then project the numeri-
cal solution onto the manifold which is determined by the con-
In this paper, we consider the numerical solutions of the follow-
ing system of nonlinear dynamical systems served quantities (2). The methods require to solve a system of
nonlinear equations at each step. The projection method is implicit
_ ¼ f ðxðtÞÞ
xðtÞ (1) even if an explicit RK scheme is applied [10]. Hence, the compu-
tational effort is large to implement projection RK methods. In
where f : Rd ! Rd is continuously differentiable, xðtÞ 2 Rd is Ref. [11], the discrete gradient (DG) methods are applied to sys-
state variable vector, and the initial vector x(0) ¼ x0 is known. We tem (1) with the conserved quantities (2). In Ref. [12], the Hamil-
always assume that system (1) has a unique solution. Furthermore, tonian Boundary value methods (HBVMs) are studied for
assume that system (1) has m conserved quantities (constants of preserving conserved quantities. Recently, Lie-group integration
motion or first integrals) [1], Ei ðxðtÞÞ, for i ¼ 1, 2,…, m scheme is discussed in Ref. [13] for preserving the angular
momentum and energy. All the above methods are implicit for
8
< E1 ðxðtÞÞ ¼ E1 ðxð0ÞÞ ¼ c1 general nonlinear system (1) with (2). For the infinite dimensional
… (2) Hamiltonian systems, the multisymplectic method is presented in
: Ref. [14]. Recently, a generalized multisymplectic method for the
Em ðxðtÞÞ ¼ Em ðxð0ÞÞ ¼ cm
infinite dimensional Hamiltonian systems with weak damping has
i.e., the m functions Ei(x(t)) are conservative for any t 0. been developed in Refs. [15] and [16]. Some explicit schemes for
The nonlinear dynamical system (1) with (2) appears in many structure-preserving have been presented in Refs. [16–20].
problems in science and engineering, for instance, mechanical With respect to the computational effort, explicit RK methods
vibration, celestial mechanics, and Hamiltonian systems [1–4]. are superior to implicit ones. However, the standard explicit
When the dynamical system (1) has only one conserved quan- RK methods do not preserve the conserved quantities well. For
tity, it is a special case of system (1) with (2). A numerical method general nonlinear system (1) with conserved quantities (2), it is
that preserves the conserved quantity is usually called the energy- impossible for any explicit RK methods to exactly preserve
preserving method. The energy-preserving method is one of the the conserved quantities [1,5,7]. In order to devise explicit
geometric numerical integration methods [5,6]. All Runge–Kutta RK methods for preserving the conserved quantities (2) of
(RK) methods preserve arbitrary linear invariants [5], and some general nonlinear system (1), the requirement Ei ðxn Þ ¼ Ei ðxð0ÞÞ
(the symplectic) RK methods preserve arbitrary quadratic invari- for i ¼ 1, 2,…, m should be relaxed.
ants [1,5]. However, no RK methods can preserve arbitrary poly- The presented paper is an extension of Ref. [6] for the case of
nomial invariants of 3 deg or higher [7]. multiple conserved quantities. We emphasize that we do not
In order to numerically preserve the energy of nonlinear require Ei ðxn Þ ¼ Ei ðxð0ÞÞ exactly holds for i ¼ 1, 2,…, but approx-
dynamical systems, many numerical methods have been devel- imately holds when an explicit RK method is applied to general
oped [1,5,8,9]. Projection methods are developed for the numeri- nonlinear system (1) with (2). We study a numerical method that
cal solutions of differential equations on manifolds [10]. The provides an approximation to the solution of system (1) preserving
methods are also applied to dynamical system (1) with (2). the conserved quantities (2) in the sense that Ei(xn) approximates
The basic principle of the projection RK methods for preserving- to Ei(x(0)) for i ¼ 1, 2,…, m.
energy is as follows. An arbitrary RK method is applied to This paper is organized as follows. A modified version of
explicit RK methods is presented in Sec. 2. In Sec. 3, we compare
the modified version of the explicit RK methods with standard RK
Contributed by the Design Engineering Division of ASME for publication in the
JOURNAL OF COMPUTATIONAL AND NONLINEAR DYNAMICS. Manuscript received
method, the discrete gradient method, and the standard projection
November 28, 2016; final manuscript received May 10, 2017; published online July method through numerical experiments. Conclusions are given in
12, 2017. Assoc. Editor: Haiyan Hu. Sec. 4.
Here h stands for a step-size and tn ¼ tn–1 þ h. When aij ¼ 0 for holds.
i j, the RK method is explicit. Hereafter, we will only consider Proof. By formula (9)
explicit RK methods.
The aim of numerical computation is to seek a numerical xnþ1 ¼ xnþ1 hpþ1 arEðxnþ1 Þ (11)
solution which is as close to the exact solution as possible.
Because a numerical solution is approximated, it is not necessary x nþ1 Þ at xnþ1 and (11), we
According to Taylor expansion of Eð
for the conserved quantities of the numerical solution to be pre- have
served exactly. It is reasonable that the conserved quantities of the
numerical solution are also approximated. In this paper, based on x nþ1 Þ ¼ Eðxnþ1 Þ a½rEðxnþ1 ÞT ½rEðxnþ1 Þhpþ1 þ Oðh2ðpþ1Þ Þ
Eð
the idea, a modified version of explicit RK methods is presented. (12)
Assume that system (1) has m conserved quantities, Ei(x(t)),
which mean that Eq. (4) holds. We want to seek a modified ver- Since rE(xnþ1) 6¼ 0 and h is sufficiently small, we obtain
sion of explicit RK methods which generates a numerical solution
xn such that E(xn) is as close to E(x(0)) as possible. In addition, x nþ1 Þ < Eðxnþ1 Þ
Eð
the order of the modified version is the same as the standard RK
method. Since we are only concerned with explicit RK methods, it The proof is completed.
is emphasized that we only require that E(xn) ¼ E(x(0)) approxi- Remark 2.4. Inequality (10) in Theorem 2.1 shows that the
mately holds. modified RK methods (8) and (9) for system (1) are superior in
The following definition describes how E(xn) is closed to preserving m conserved quantities to the standard one (8).
E(x(0)). The following theorem is concerned with the order of the modi-
DEFINITION 2.1. Assume that a given positive constant e is suffi- fied Runge–Kutta method.
ciently small. An explicit numerical method is called e-preserving THEOREM 2.2. Assume that the localizing assumption xn ¼ x(tn)
if it generates a numerical solution xn such that holds. If ajjrEðxnþ1 Þjj is bounded, then, the order of the modified
Acknowledgment
The author wishes to thank the two anonymous reviewers for
their valuable comments and suggestions which are helpful to
improve this paper. One of the reviewers let the author know
some recent references. This work was supported by the National
Natural Science Foundation of China (Grant Nos. 11371053 and
61372090).