本計劃主要是在多處理器上探討二或三維度拋物線型偏微分方程式的數值方法。有線差分法為其主軸，其中包刮了交錯方向隱性法及交錯芳現險性法。在多處理器上用交錯方向隱性法求解，必定會遭遇到要使用高思消去法、空間分割、及資料交換等問題，這些都是我們要去一一克服的問題。目前我們為週期邊界問題設計一個平行算法。如果能善用區域資料和適當的資料傳遞，可以提升交錯方向隱性法的效率。 We investigate the implementation of several classical methods for solving parabolic equations for higher (2 or 3) dimensions on multiprocessors. The methods considered are the Alternating Direction Implicit (ADI) algorithms and the Althmating Direction Explicit (ADE) methods. We focus on parallel ADI methods which coontains Gaussian elimination, domain decomposition, and data communication. Currently, we develope a algorithm for periodic problems. The high efficiency of the method is achieved by carefully using local data and suitable data passing.
