applied to a linear PDE (ut + ux = 0). Moreover, for some Runge–Kuttamethods, if one looks at the intermediate stages, e.g. u(1) (1.8) or (1.9),one observes even bigger oscillations. Such oscillations may render difficultieswhen physical problems are solved, such as the appearance of negativedensity and pressure for Euler equations of gas dynamics. On the otherhand, SSP Runge–Kutta methods guarantee that each intermediate stagesolution is also TVD.Another example demonstrating the advantage of SSP methods versusnon-SSP methods for shock wave calculations can be found in [31].Further examples comparing some commonly used non-SSP methods withSSP methods may be found in, e.g., [62, 55, 54]. These numerical examplesdemonstrate that it is at least safer to use SSP time discretizationswhenever possible, especially when solving hyperbolic PDEs with shocks,which make the traditional linear stability analysis inadequate. In termsof computational cost, most SSP methods are of the same form and havethe same cost as traditional ODE solvers. It is true that the time step tmight need to be smaller to prove the SSP property than, say, when linearstability is proven, however in many situations t can be taken larger inpractical calculations without causing instability.Development of SSP methods was historically motivated in two ways,and developed by two groups: one focusing on hyperbolic partial differentialequations, the other focusing on ordinary differential equations. Manyterms have been used to describe what we refer to as strong stability preservation;here we stick mostly to this term for clarity.
đang được dịch, vui lòng đợi..