The initialization of equation-based differential-algebraic system models, and more in general the solution of many engineering and scientific problems, require the solution of systems of nonlinear equations. Newton-Raphson's method is widely used for this purpose; it is very efficient in the computation of the solution if the initial guess is close enough to it, but it can fail otherwise. In this paper, several criteria are introduced to analyze the influence of the initial guess on the evolution of Newton-Raphson's algorithm and to identify which initial guesses need to be improved in case of convergence failure. In particular, indicators based on first and second derivatives of the residual function are introduced, whose values allow to assess how much the initial guess of each variable can be responsible for the convergence failure. The use of such criteria, which are based on rigorously proven results, is successfully demonstrated in three exemplary test cases.
翻译:以方程式为基础的差异代数系统模型的初始化,更一般地说,是许多工程和科学问题的解决办法,都需要非线性方程系统的解决办法。 牛顿-拉夫森的方法为此广泛使用;如果最初的猜想足够接近它,在计算解决办法时效率很高,但有可能失败。本文采用了若干标准,分析最初的猜想对牛顿-拉夫森算法演变的影响,并确定在趋同失败时哪些初步猜想需要改进。特别是,引入了基于留置函数第一和第二衍生物的指标,这些数值能够评估每个变量的最初猜想对趋同失败可能承担多大责任。在三个示范性测试案例中成功地展示了基于严格证明的结果的这类标准的使用。