高等数值算法与应用十（十一）-wenjianyu.pdf

高等数值算法与应高等数值算法与应用 （十十一） AdvancedAdvanced NumericalNumerical Alggorithms Apppplications 计算机科学与技术系计算机科学与技术系 喻文健喻文健 LectureLecture 1010 – ODEODE--IVPIVP TechniquesTechniques Two implicit methods Backward Euler Disadvantage: solve algebraic equation to get y k+1 Advantage: unconditionally stable 对稳定的问题，无论取怎样 的步长h，求解过程都稳定 Trappezoid ((二阶准确度)) Stiffness, stiff pproblem Stability:Stability: rapidrapid componentcomponent Two restrictions for step size Accuracy: slow component Some numerical methods are inefficient because of severe stability limitation Factors: ODE;; numerical method;; initial conditions 2010-12-09 Wenjian Yu 2 LectureLecture 1010 – ODEODE--IVPIVP TechniquesTechniques Runge-Kutta method Single-step, need not compute higher derivatives Order-2 scheme Order-4 scheme Self-startingg; automatic Rungge-Kutta solver Multistep method 线性多步线性多步 减少隐格式计算：PECE scheme (predict, eval, correct, eval) OrderOrder-44 AdamsAdams PECEPECE, backwardbackward differentiationdifferentiation formulasformulas 局部截断误差阶与稳定区间 不容易更改步长 也可对模型也可对模型 设等号右边项准确，y =y(t )，计算l =O(hp+1) k k k+1 问题推导 推导误差增长因子e =(?)e +O(hp+1), |(?)|≤1 kk+1 kk 对模型问题(λ为实数/复数)推导，得到稳定阈值，即hλ≥? 2010-12-09 Wenjian Yu 3 MMatltlabb iissues Matlab commands for ODE nonstiff:nonstiff: ode45ode45 (Runge(Runge-Kutt