保持梯度系统能量耗散率的松弛Runge-Kutta方法
Energy dissipation rate preserving relaxation Runge-Kutta methods for gradient systems
摘要
[目的]梯度系统沿能量下降方向演化,在连续情形下其能量耗散量可由能量等式精确刻画,其导数即为能量耗散率;基于保结构思想,构造能精确保持离散能量等式(从而精确保持能量耗散率)的数值格式至关重要,但当势函数为一般形式时,此类格式的构造面临显著困难.该文旨在发展一种能严格保持离散能量等式的数值方法.[方法]首先指出,标准Runge-Kutta方法对具有一般形式势函数的梯度流方程,在离散水平上无法精确保持离散能量等式;随后借助松弛技术,提出一类松弛Runge-Kutta 数值方法,并从理论上证明该方法具有与原Runge-Kutta方法相同的精度.[结果]数值实验结果表明,该松弛Runge-Kutta方法能够精确保持离散能量等式,验证了其精度与有效性.[结论]该方法为一般形式势函数下的梯度流方程提供了一种能严格维持离散能量耗散特性的有效保结构数值方法.
[Objective]Gradient systems evolve along the direction of decreasing energy.In the continuous setting,the energy dissipation over any time interval can be precisely described by an energy equality,and its derivative yields the energy dissipation rate.Guided by the structure-preserving philosophy,it is essential to construct numerical schemes that preserve exactly the discrete energy equality(and hence the energy dissipation rate).When the potential function takes a general form,however,the construction of such schemes encounters significant difficulties.This paper aims to develop a numerical algorithm capable of strictly preserving the discrete energy equality.[Method]It is first pointed out that the standard Runge-Kutta method fails to preserve exactly the discrete energy equality at the discrete level for gradient flow equations with a general potential function.By invoking the relaxation technique,we then propose a class of relaxed Runge-Kutta numerical algorithms and prove theoretically that they possess the same accuracy as the original Runge-Kutta method.[Result]Numerical experiments demonstrate that the proposed relaxed Runge-Kutta algorithm can exactly preserve the discrete energy equality,confirming its accuracy and effectiveness.[Conclusion]For gradient flow equations with a general potential function,the proposed algorithm provides an effective structure-preserving numerical method that can strictly maintain the discrete energy dissipation property.
年,卷(期):
2026, 48(4)
分类号:
O241.81
机标分类号:
O175
O241.82
TP391.41
在线出版日期:
2026年9月22日
基金项目:
12271463:国家自然科学基金,